Subscribe now

The State of LLMs, Part I: Building Blocks

For years, the simplest way to describe AI progress was bigger models, more data, more compute. That formula helped usher in the LLM era, but it no longer tells the whole story.

This is the first part of a two-part conversation on where AI and large language models are today and where they may be heading. I sat down with two of my favourite people in the AI community, Christos Baziotis of Cohere and Christos Perivolaropoulos of Google DeepMind.

Part I starts with the building blocks; Part II turns to agents, harnesses, benchmarks, memory, and what today’s AI systems still cannot do.

In Part I, we cover four questions:

1. How Christos Baziotis and Christos Perivolaropoulos found their way into two of the world’s leading AI companies.

2. What transformers actually do, and why they became the architecture behind most modern AI systems.

3. Why bigger models are no longer the whole story, even though scaling still matters.

4. What post-training actually does, and whether it teaches models new abilities or helps them use capabilities they already had more reliably.

Let’s get to it.

Their paths into AI

So how’d you get into AI? I’d love to hear about the moments along the way that brought you to the problems and technologies you work on today.

C. Perivolaropoulos: When I was in school, I was active on PC Magazine’s forum, a large computer magazine at the time, where a small group of members pushed each other to code and enter programming competitions. That was probably the beginning.

I studied Electrical and Computer Engineering at the University of Patras, and while at university, I worked at the Greek chip design company, ThinkSilicon, long before its acquisition by the US semiconductor leader Applied Materials. There, I spent about three years working on compilers and assemblers, the low-level systems environment that converts human-readable code into binary machine code, 0s and 1s, that a computer can execute. This shaped a lot of how I think about performance and hardware.

While still at university, I began collaborating with MIT as a Visiting Researcher and worked at another Patras-based startup, Codebender, which built developer tools for programming Arduino-style hardware. Eventually, I decided to pursue a PhD at the University of Edinburgh in 2016, focusing on query languages and databases, and interned at Google and Amazon.

After the PhD, I spent some time at Huawei and later joined DeepMind, where I am today. My work is split between making AI models run more efficiently on GPUs and studying how models, mainly LLMs, but also graph models and other systems, can or cannot generalise beyond their training data.

C. Baziotis: I studied Computer Science at the University of Piraeus. That was where I first saw traditional AI, such as rule-based systems and evolutionary algorithms. I liked it, but I did not imagine I would actually work on AI.

After graduating, I worked as a software engineer in Greece while doing a master’s degree in the same department. During that master’s program, I discovered modern machine learning and natural language processing, a field focused on helping computers understand and generate language. For my thesis, I participated in SemEval, an international competition for language-processing systems, and our work topped the leaderboard. That was the first time I felt I could maybe do research seriously.

I started a PhD at NTUA in 2016. The lab was great, but the field was moving extremely fast, so I decided to leave a year and a half later and restart my PhD at the University of Edinburgh, focusing on pre-training, the stage where a model first learns broad patterns from large amounts of data, and synthetic data for low-resource machine translation. I also interned at Meta and Amazon.

Then, in 2023, I joined the founding team of Samaya, an AI platform for finance, where I worked on the search and agentic stack. The team was strong, and it was a lot of fun. But after two years, I started to miss working on frontier research and LLMs, so I joined Cohere. I now work mainly on memory for agents and on helping models learn from experience after deployment.

The architecture behind LLMs

If you had to explain transformers to an outsider, what is a transformer actually doing? As we ask these models to do more complex things, do you think the basic transformer idea is starting to show its limits?

C. Perivolaropoulos: A transformer is an architecture that helps an AI model connect the relevant pieces of information together.

Before a transformer can do that, the model has to turn information into a form it can work with. Words, ideas, documents, or pieces of information may look separate to us, but neural networks turn them into mathematical representations called embeddings. Embeddings are vectors, lists of numbers that act a bit like coordinates in a model’s learned map of meaning. Related concepts tend to sit closer together in that space, while dissimilar concepts tend to sit farther apart.

The transformer’s key move is association. It looks at the information you give it and determines which pieces are relevant to one another. If I give it a set of files or some information, the transformer can connect the relevant parts of that input to one another and to patterns it learned during training.

Through many layers of this process, the model builds increasingly rich representations. In an LLM, those representations are used to predict the next token. In another kind of AI system, they might be used to generate an image, etc.

So, at a high level, a transformer is not just memorising facts. It is building relationships between pieces of information and using those relationships to generate something useful. At that level, I do not think the basic idea is breaking down. Transformer-like architectures are still evolving and producing gains.

The limits show up when we ask models to generalise beyond what they have learned. One example is length generalisation. If a model learns a small version of a task, can it scale that task up? If it learns to multiply three-, four-, or five-digit numbers, does it then know how to multiply fifty-digit numbers? With current architectures, the answer is no.

But when you look closely, the failure often comes from specific design choices in today’s models, such as how they handle position or how they choose between possible outputs. It is not necessarily that the core idea of connecting tokens to one another is fundamentally broken.

The current architecture has real limitations, but I believe we can keep modifying and improving transformer-like systems and continue to get gains.

C. Baziotis: To explain transformers, it would be useful to compare them with the models that came before them. Before transformers, many language models worked more like someone reading a text line by line. They would read the first word, update an internal summary, then read the next word and update that summary as they went. These models were called RNNs, or recurrent neural networks.

That approach made intuitive sense because language is sequential, with one word following another. But it had two big problems. First, the model had to squeeze everything it had read into one running summary. If something important appeared early in a long paragraph, it could get lost by the time the model reached the end. Second, the model had to process the text step by step, which made it hard to train efficiently on GPUs.

Transformers changed the approach. Instead of reading one word at a time, a transformer looks at a whole window of text at once. That window is called the context. Within that context, the model uses attention, a mechanism that helps each word find the other words or pieces of information that matter to it.

You can think of attention as a shortcut. The model no longer needs to carry everything in a single running summary. If a word near the end of a paragraph depends on something near the beginning, it can connect to it directly. That was the breakthrough. Transformers could train much faster, use GPUs much better, and learn from much larger datasets. They also became much better at connecting information across longer texts.

But that design also creates the main limitation. A transformer only works with what is inside its current context window. If the information is inside the window, the model can use it. If not, the model does not naturally have access to it. This is why so much work in AI has focused on increasing context windows and managing them more effectively. But context is not free. The more information you put into the window, the more expensive and noisy the process becomes.

Longer context helps, but it is not the same thing as memory. Memory means the system can keep track of useful information over time, update what it knows from experience, and carry that forward into future tasks. A larger context window just means the model can see more at once.

Overall, I would not say transformers are breaking down. They solved a very important problem and remain extremely powerful. However, their limits are becoming clearer as we ask AI systems to behave less like chatbots and more like long-running agents.

The likely future is not that we throw transformers away. It is that we keep what works, such as attention, parallel processing, and scalable training, while adding better ways for AI systems to remember, update, and use information over time.

Scaling laws

For a while, the story in AI was bigger models, more data, better results. Is that still true, or is the field starting to run into limits where model size alone is no longer enough?

C. Baziotis: The short answer is that model size still matters, but it is no longer the whole story. For a long time, the field had a very powerful recipe: make the model bigger, train it on more data, use more compute, and performance tends to improve. The technical term behind that intuition is scaling laws.

A scaling law is an observed relationship between a resource and an outcome. The resource could be the number of parameters in a model, the amount of training data, the amount of compute used during training, or even the amount of compute used when the model is answering a question.

In the early LLM era, scaling mostly meant pre-training larger models on more data. The basic idea was that if you make the model bigger and give it more data, performance improves in a fairly predictable way. That predictability is what made scaling laws so important. They did not just say “bigger is better”; they gave labs a way to estimate how much better a model might become by adding more data, more parameters, or more compute.

That confidence was slightly shaken around GPT-4. GPT-4 was OpenAI’s major model released after GPT-3.5, the model family behind the first version of ChatGPT. OpenAI has not publicly disclosed the exact parameter counts for GPT-4, but it was definitely larger and clearly more capable than GPT-3.5. Still, many people expected an even more dramatic leap, partly because it was rumoured to be much larger. That led to speculation that model-size scaling was starting to hit diminishing returns.

I do not believe the right conclusion is that scaling laws are dead, but rather that scaling has become more complicated. You can’t just keep making the model bigger and expect the same returns forever. If you increase the model size, you also need enough high-quality data to train it. If high-quality data becomes harder to find, as it increasingly seems to be, then model size alone is no longer the obvious path forward.

That is why the industry is now exploring other forms of scaling, such as test-time compute, which allows the model to use more computation when answering difficult questions. This is what made reasoning models such as OpenAI’s o1 so interesting. The model can “think” for longer, and in some cases, that improves the answer.

Another important axis is post-training, the stage after a model’s initial training, where it is refined to behave more usefully. One important post-training method is reinforcement learning.

Instead of only asking, “How big should the model be?”, labs now ask, “How much should we spend shaping the model after pre-training? How many examples, rollouts, or feedback loops should we use? How much performance do we get from that?” So the question today is not simply whether larger models still improve. They often do. The question is whether making the model larger is the best use of resources.

When people say a scaling law is “breaking,” they may mean different things. It could mean the model can no longer learn much more, or that we do not have enough good data, or that the data exists but is too expensive to collect. Or it could mean that spending more compute yields diminishing returns.

Scaling is still a useful guide, but the simple version of the story is over. The field is moving from “make the model bigger” to “find the best axis to scale.” Right now, post-training with reinforcement learning looks like one of the most promising areas to scale.

C. Perivolaropoulos: I agree that scaling laws remain a fundamental guide, but their application is changing because the landscape of data and compute is changing. We are still seeing substantial returns from scaling, even if those are no longer exponential. The transformer has proven highly scalable, but it is also resource-intensive. In earlier stages, the field benefited from an abundance of data and compute. Today, there is a shift toward efficiency.

As high-quality data becomes scarcer, the challenge is less a simple breakdown of scaling logic and more a logistical and economic constraint: where do you get the data, what does it cost, and how do you use the compute most effectively? In that sense, scaling laws still matter, but the relevant bottleneck may be the practical cost of moving further along the curve.

That is why efficiency has become so important. The field is no longer only asking whether bigger models work; it is asking how to get more out of each unit of data and compute.

How models improve after initial training

We talked about post-training and reinforcement learning as new ways labs improve models after the initial training run. But what are these methods actually doing? Are they teaching models genuinely new abilities, or helping them use capabilities they already have more reliably?

C. Perivolaropoulos: Post-training does something important, but in a sense, it sits outside the base model. One uncharitable view is that LLMs have fundamental flaws, and post-training is a set of bandages that merely cover them. There is some truth to that, as these methods work partly because the base model is not perfect.

But I do not like that framing. Post-training is better understood as another way to turn compute, data, and feedback into more useful behaviour. It is not always about making the model fundamentally smarter in some direct way. It is about shaping the system so that the model produces better, more reliable outputs.

That is why I see post-training as part of a broader process for extracting useful behaviour from probabilistic models. The same general idea may not be specific to LLMs; similar approaches could potentially be adapted to other types of models as well.

C. Baziotis: My view is that many gains from post-training today come from elicitation. By elicitation, I mean that the model already has many capabilities from pre-training, and post-training helps bring them out more reliably.

You can see this in areas such as maths and code. In the first few steps of post-training, performance can improve very quickly. That probably does not mean the model suddenly learned an entirely new skill. More often, post-training makes the model more likely to produce the right format, follow a useful strategy, or apply knowledge it already has.

There is evidence pointing in that direction. For example, some papers last year found that Qwen2.5 could improve on maths even when using reward signals that were random or explicitly incorrect. That suggests the training process may be amplifying behaviours that were already latent in the model, rather than teaching a completely new capability from scratch.

Papers have shown a similar pattern when sampling many answers from a base model. Even before post-training, the correct answer may already appear in some fraction of cases. Post-training changes the model’s behaviour so that the good answer comes out more reliably.

That also explains why progress has been strongest in domains such as maths and code, where correctness is easier to verify. If you can check whether an answer is right, it is easier to give the model useful feedback. In messier real-world domains, where correctness is more subjective or harder to measure, post-training becomes much harder.

Share

—

Part I leaves us with the conclusion that the old story of AI progress is no longer enough. Bigger models still matter, but the real gains now come from how these models are built, scaled, and refined. Part II turns to what they can actually do. Once a model is powerful enough, the harder questions become practical ones: Can it use tools? Can it remember what came before? Can it work through long tasks without losing the thread? And can we even measure whether it’s useful?

That’s where we go in State of LLMs: Part II in two weeks. Stay tuned.

Top News

We invested in NOFire AI to help engineering teams prevent, resolve, and automate responses to production issues

AI is helping software teams write and ship code faster than ever. But when more code reaches production faster, things can also break faster, and finding the real cause of an outage can feel like searching for a needle in a haystack.

Founded by Spiros Economakis, Panagiotis Moustafellos, Anastassios Nanos and Antonios Chalkiopoulos (ex Elastic, Mattermost, Lenses), NOFire AI helps engineering teams understand what changed, what broke, and why. It connects the dots across a company’s software systems, identifies risky updates before they cause problems, and helps teams resolve incidents in minutes instead of hours.

We led NOFire AI’s $2.5M Seed round to build the reliability layer for the age of AI-generated software.

New drug could be a game-changer for people with high cholesterol

Eli Lilly reported a few days ago that a high dose of its experimental gene-editing therapy cut LDL cholesterol by 62% in an early clinical trial, a promising sign that a single treatment might one day replace the daily pills people now take to manage their cholesterol. Unlike statins, the therapy, called VERVE-102, works by making a one-time edit to a gene in the liver, so its effect is meant to last for years rather than wearing off between doses.

Lilly gained the treatment through its roughly $1B acquisition last year of Verve Therapeutics, a startup whose origins trace back to physician and scientist Anthony Philippakis, who spearheaded its founding at Google Ventures in 2016 to bring genetic medicine to the treatment of heart disease.

How well do LLMs understand and perform in Greek?

What happens when you throw LLMs into messy, high-context Greek scenarios with ambiguity, bureaucracy, domain-specific language, outdated public information, false assumptions and local jargon? Greek LLM Arena is an open-source benchmarking platform designed to assess how well large language models understand and perform in Greek.

Jobs

Check out 320+ fresh job openings from startups hiring in Greece here.

Fundings

Drone delivery platform Matternet raised $33M and went public via a reverse merger.

August Robotics raised $30M in Series B funding, led by Big Pi Ventures, to build autonomous robot fleets for construction, industrial, and large-scale projects.

AI customer support platform Gradient Labs raised a fresh $13M in funding led by Octopus Ventures, doubling its Series A to $26M.

Twin Prime announced a $10M Pre-seed round led by Expeditions to build AI models for defence and security.

AI-powered sales training platform Deelan AI raised a Seed round led by Expon Capital.

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a ❤️ and share.

Alex

Subscribe now

The Alchemist’s Cap Table

There’s a hot sector in tech right now soaking up capital, talent, and attention. A new crop of research-first AI startups is raising extraordinary sums, often with little revenue, few customers, and sometimes no public product. Companies seem to reach billion-dollar valuations weekly, on the strength of a founding team, a technical thesis, and the belief that they’re close to the next breakthrough.

This feels new because the numbers are so large and the timelines are so compressed, but the deeper pattern is older. AI is starting to look less like the software markets of the 2010s and more like biotech in the 1980s.

In 1980, Genentech went public and became one of the great startup stories of the decade. It was a four-year-old biotechnology company built around a radical idea: use engineered cells to make human proteins that could become medicines. Genentech had only a few million dollars in revenue and no marketed drugs of its own, yet investors believed that a small biology lab could become a pharmaceutical giant.

Genentech’s IPO helped set off a biotech frenzy. Venture capital was expanding, public markets were hungry for new stories, and young labs were suddenly treated as future public companies. But the market soon realised that discovering something in a lab wasn’t the same as bringing medicine to patients. Between the lab bench and the patient stood clinical trials, regulators, factories, and salesforces. Big Pharma owned much of the machinery needed to bring a discovery to market.

By the late 1980s, the pattern was obvious enough that the Wall Street Journal described many biotech startups as “bred just to be sold”. Large pharmaceutical companies were buying young biotech labs not because they looked like mature businesses, but because they contained valuable science, platforms, and people. Bristol-Myers bought a monoclonal-antibody company that didn’t generate cash. Novo Nordisk bought a recombinant-protein lab and turned it into its discovery arm. Even Genentech eventually bent toward Big Pharma with Roche buying a controlling stake for $2.1B in 1990.

The biotech boom was real. Some labs eventually brought their own drugs to market and became lasting companies. But biotech also taught investors that a lab could be highly valuable even before it was a viable enterprise. The startup gave a molecule, an assay, a platform, or a research team a corporate boundary, and Big Pharma could then decide whether to partner with it, fund it, buy it, or absorb it.

Software had versions of this, too. But for most of the SaaS, mobile, and social eras, the startup was usually built around a product, a customer workflow, a distribution wedge, or a user network. The scarce thing was adoption. Had a small team found a new habit before the platform did? Had a workflow embedded itself deeply enough in a customer’s day? Had a network started compounding?

AI has brought the lab back to the centre of gravity.

OpenAI, then Anthropic, made the category believable. They showed that an AI lab could become one of the most valuable companies in the world, not merely a research division inside Google, Meta, or Microsoft. But their success created the same temptation that Genentech created forty years earlier. Once a few labs became the companies everyone pointed to, the market began treating many more as if they were next in line.

People have started calling these companies “AI neolabs”. They’re private research companies, usually founded by well-known researchers or repeat founders, that raise huge amounts of money before they have much revenue, or sometimes even before they have a public product.

Investors aren’t betting on current sales but rather on the team’s proximity to a technical breakthrough. Thinking Machines Lab is the cleanest example. Founded by former OpenAI CTO Mira Murati, it reportedly raised $2B at a $12B valuation, the largest Seed round in Silicon Valley history, before releasing a public product. Or Reflection AI, founded by ex-DeepMind researchers Ioannis Antonoglou and Misha Laskin, which raised billions to build advanced open source models that could challenge China’s DeepSeek.

Like early biotech companies, AI neolabs are built around rare technical talent, expensive experiments, and know-how that’s hard to write down. That makes them hard to value like normal software companies. Their value isn’t always concentrated in a single finished product, but rather spread across founders, researchers, models, data, product, and lessons learned from failed experiments.

The company gives that work a boundary, but not necessarily its final home. Adept showed one version of this. The company raised $415M to develop AI agents. In 2024, Amazon reportedly spent more than $300M to license Adept’s technology and hire some employees, including its CEO, while leaving the company technically independent under new leadership. Inflection followed a similar path. After raising $1.5B to build a consumer-facing AI assistant, Microsoft hired much of the team and paid $650M, largely to license its models. Neither of them had significant revenue streams at the time.

These deals looked strange because they weren’t clean acquisitions. They were reverse-acquihires where the buyer didn’t simply buy the whole company. Instead, it hired key people, licensed the technology, and left the corporate shell behind. The startup may technically remain independent, but the centre of gravity has shifted, often leaving the remaining staff holding a bag of options that are definitely not worth what they should be.

The biotech startup of the 1980s was often bred to be sold. The AI neolab of the 2020s is built to be unbundled. Not necessarily by intent, but by structure.

Two forces collide here. First, the assets needed to scale AI are expensive and concentrated. AI neolabs need compute, distribution, and capital that sit disproportionately with Big Tech. A small lab can discover a capability long before it has the machinery to bring it to the world. Second, Big Tech cannot simply acquire every strategically important AI company without inviting scrutiny. So the deal changes shape.

The biotech analogy cuts both ways. The 1980s biotech boom produced hype, overfunding, and companies that never became standalone businesses. But it also produced real science, real medicines, and enduring companies. AI will likely rhyme, as in, a few neolabs will become important independent companies, while many others will disappear, sell, or be unbundled into larger platforms.

In a narrow sense, the AI neolab has something of the old alchemist’s workshop. Not the mysticism, but the structure: expensive instruments, secret recipes, patronage, failed experiments, and the hope that enough attempts will reveal a method the powerful cannot afford to leave outside.

Tracing back to as early as Greco-Roman Egypt, alchemists sought to transform matter and uncover hidden laws of nature. The field attracted frauds, but also produced serious experimenters such as Paracelsus and Jan Baptist van Helmont, whose work helped lay the groundwork for chemistry and medicine. And, as the powerful institutions of their time, rulers gathered alchemists around their palaces not because they knew which experiment would work, but because they wanted access to the people who might discover something valuable.

Today’s workshops have GPUs instead of furnaces and model weights and reinforcement learning techniques instead of alchemical recipes. Around them gather venture capital, cloud providers, and Big Tech platforms, not monarchs and nobles. But the old logic remains:

The palace doesn’t always know which experiment will matter.

It only knows that it wants the people who might find out.

Share

Top News

Blackstone acquired a majority stake in Skroutz

The world’s largest alternative asset manager, Blackstone, acquired a majority stake in Greece’s leading e-commerce platform. Apart from the private equity fund CVC Capital Partners, which is divesting its position in the company, Skroutz’s founders are set to divest a portion of their shareholding while retaining a stake and continuing to lead the business. George Hadjigeorgiou is remaining as CEO. According to Reuters, the deal values ​the Greek company at ​about €635M.

London Greeks In Tech on June 11

Come join us for a Greeks In Tech event in Central London, right near Farringdon Station, on June 11. You can RSVP here.

Train your own LLM from scratch

A hands-on workshop from Angelos Perivolaropoulos at ElevenLabs, explaining how to train a language model locally yourself, with a practical focus on the tooling, constraints, and engineering tradeoffs. Links from YouTube and GitHub.

Summer Founders Program by BuildAthens

A founders program running July 1 – September 30 in Athens and organised by Achilleas Mitrotasios and Thomas Kosmas, offering $20K in funding, office hours and support, dinners with founders, and intros to investors. You can apply by May 31.

Jobs

Check out hundreds of job openings here from startups hiring in Greece.

Fundings

AI-powered primary care doctor Clara raised $12M Pre-seed led by YCombinator and A.Capital.

Neurosoft Bioelectronics raised $7.5M Seed led by Skybound to develop brain-computer interface technology.

Cancer diagnostics company Imagenomix secured $3.2M in Seed funding led by Modi Ventures, with participation from Metavallon VC.

Agentic cybersecurity Hoplon raised $1M Seed led by Haatch and Genesis Ventures.

Neptune Zero announced a €500K round led by Corallia Ventures for compliance and decarbonisation in marine operations.

E-mobility startup FlexThis raised €150K by Investing for Purpose.

AMCO raised funding from Halcyon Equity Partners to develop technology in intelligent transport systems, micromobility, and smart cities.

Perceptual Robotics raised funding from Loggerhead Ventures and One Planet Capital for automated wind turbine inspections.

Healthcare companies Health4Crew and Puberry raised funding from Envolve.

Acquisitions

Leading Japanese semiconductor manufacturer Renesas Electronics acquired Patras-based Irida Labs, a company developing edge computer-vision software that transforms cameras into intelligent AIoT sensors.

New Funds

Skybound launched with $38M to back deep tech founders in Pre-Seed and Seed.

Starcinco is a new defence tech fund with €22M focusing on Greece, Cyprus, and Armenia.

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a ❤️ and share.

Alex

We’ve had guests working on foundation models, humanoids, space rockets, new drugs, and nuclear fusion. Founders who raised billions, scaled to millions of users, sold their businesses, pivoted, or stayed bootstrapped.

Thank you to everyone who’s been part of it until now. I hope this has helped even a few of you connect the dots and go pursue your craziest, most ambitious ideas.

If you’re not a subscriber, click below to get the next posts in your inbox:

Subscribe now

On April 1, 2026, Artemis II launched from Kennedy Space Centre, carrying four astronauts aboard its crew spacecraft Orion on humanity’s first crewed lunar flyby in more than fifty years. Over roughly ten days, the crew tested Orion’s systems in deep space, travelled farther from Earth than any humans have ever gone, swung around the Moon, observed its near and far sides, and splashed down in the Pacific on April 10.

Artemis II brought back a storyline that had been dormant for more than half a century: humanity is back in lunar space. We didn’t land on the Moon, as that was never the mission. But the Moon has re-entered the centre of politics, national strategy, and long-range industrial planning. This time, the story isn’t just about getting back to the Moon. It’s about what happens once we do.

That is the harder part to understand. Over the past few weeks, Dimitris Anastasiou and I have been talking through what Artemis II really changed, why the Moon is suddenly urgent again, and why the next chapter of space may depend less on rockets than on robots. Dimitris is a Founding Engineer at Icarus Robotics, where he’s building robotic labour systems for space. He thinks about the Moon less as a destination than as a workplace under construction. This essay is our attempt to make sense of what that means.

Let’s get to it.

Who Will Build The Moon?

A co-written essay with Dimitris Anastasiou

Artemis II put the Moon back in motion

For more than half a century, the Moon belonged mostly to history. Apollo 17 left the lunar surface in December 1972, and no human being returned. The Moon remained culturally powerful, technologically relevant, and strategically tempting, but dormant as a real theatre of human action. Artemis II changed that, not because it completed humanity’s return, but because it proved that “return” is now a live campaign.

That campaign is easy to misunderstand if we see it through the lens of Apollo nostalgia. The reason the Moon faded after Apollo was that the US had already won the original space race. Once that symbolic victory was secured, the political incentive weakened, budgets tightened, and lunar missions ceased to be a national priority.

This also helps explain why the Moon is back now. The United States is approaching this moment as a strategic priority. At this year’s Space Symposium, the NASA administration made clear that the US needs to win the space race against China. Whether you take that as rhetoric, doctrine, or both, the point is that the space race is back because the incentive structure is back. Meanwhile, the sense is that China is moving faster than many expected, even if much of its progress remains opaque.

The ecosystem is very different now from what it was in the 1960s. NASA is no longer the only meaningful actor. SpaceX, Blue Origin, Voyager, and a wider class of commercial companies now shape the field too. The space industry is commercialising, but the initial push remains political.

Which means we should read missions like Artemis II (which was designed to validate systems, procedures, manual operations, life support, navigation, and the broader machinery of sending humans into deep space), less as a lunar spectacle than a systems check on a geopolitical ambition.

And yet, even at this stage, we should be sceptical of the more inflated visions. People like to imagine lunar cities, giant domes, and self-sustaining bases. The first wave of real lunar infrastructure will look nothing like that. It will be smaller, rougher, and far more dependent on Earth for logistics, resupply, navigation, and maintenance. It will look more like an Antarctic research outpost than a science-fiction metropolis.

Artemis II forces a harder question: What happens after the mission patches, livestreams, and national statements fade? What, exactly, does it take to build something that lasts?

Rockets get headlines. Labour builds infrastructure.

Space’s real bottleneck is labour. We’re used to talking about space in terms of launch cadence, propulsion, and rocket architecture. Those things matter. But if the ambition is not just to visit space, but to build there, then the question changes to: Who does the work?

If we want functioning stations, orbital logistics, maintenance, inspection, servicing, experimentation, and eventually larger infrastructure around the Moon, we need persistent labour. This is where the usual story about astronauts stops being enough. Astronauts are too few, too expensive, and too unscalable to serve as the foundation for a real space economy. It costs $130,000 per hour to keep an astronaut in orbit. Human spaceflight can push the frontier, but it cannot, on its own, industrialise it.

That’s the opening for robotic labour. If space is going to move from heroic exploration to actual infrastructure, then a meaningful share of the repetitive, operational, and physically costly work has to become robotic.

The labour force for space

This is the problem Icarus Robotics is trying to solve: How to build robotic workers capable of performing real operational work in orbit. We’re starting with a specific kind of machine, a free-flying, dual-arm robot designed for crewed space stations. It can move through the interior of a station, float from task to task, and use two manipulators to perform actual work.

The initial target is station logistics, such as cargo handling, internal maintenance, inspection, and repetitive, low-leverage tasks that need to be done and require astronaut time. We’re starting where the problem is hard enough to matter but not so hard that every variable hits at once, like inside space stations in low Earth orbit with intravehicular applications.

Inside a station, you’re still dealing with microgravity, control complexity, manipulation, and reliability. But you aren’t yet taking on the full stack of external exposure to vacuum, extreme temperatures, and radiation as you would outside of a spacecraft. That lets us focus first on controllability and meaningful work inside the habitat.

The first mission, Joyride 1, is targeted for early 2027 with a test flight on the International Space Station (ISS) in collaboration with Voyager. More specifically, the point is to prove that a free-flying, dual-arm robot can operate inside a crewed station and do useful work there. Not just exist in orbit but work in orbit.

Right now, Icarus’s system is teleoperated by ground operators using dedicated interfaces to fly the vehicle and control its two manipulators. However, the long-term opportunity is autonomy, since it’s the only path to scaling work in space as missions become more complex and continuous.

In the future, every serious space robot will require a software layer capable of managing autonomy, reliability, and environmental adaptation. That’s the layer we want to help define with Icarus Robotics, and our long-term ambition is a software-and-autonomy layer for robotic labour across space infrastructure, from internal station logistics to satellite servicing, refuelling, maintenance, and eventually the systems that support places like future lunar bases.

Why space robotics is harder than it looks

The first challenge of developing robots for space is physical. Microgravity changes control. The coupling dynamics of robotic systems in space differ from those on Earth. Even before you reach extreme external environments, the absence of gravity changes the problem.

The second challenge is operational. On Earth, you can run tests, iterate, debug, and repair. In space, by contrast, the robot must operate in environments where support is sparse, repair is difficult, and failure is costly. That means the standard for reliability is far higher than in most terrestrial robotics environments.

The third challenge is epistemic. You cannot fully recreate the real environment on Earth. You can simulate it, approximate it, test pieces of it through air-bearing setups, neutral buoyancy, and parabolic flights, though none of those perfectly substitutes for the real thing.

So, deployment is such a meaningful threshold. The problem isn’t just getting a robot to function. It’s getting a robot to function reliably, out of the box, in an environment that resists full rehearsal. In space, there’s no clean separation between control, deployment, operations, and reliability. They collapse into each other. And that collapse is exactly why the field has to confront the difference between what can impress in a demo and what can be trusted in infrastructure.

The AI stack is still unsettled

Then, there’s the deeper question: Even if the need for robotic labour is obvious, do we actually know how to build the intelligence stack that will support it? There are three unresolved layers. The first is hardware. Compared with the human body, and especially the human hand, robotic hardware still lags. Tactile sensing, proprioception, and overall physical fidelity remain far from what humans take for granted.

The second is architecture. Robotics doesn’t yet have its transformer moment. There’s no agreed-upon stack that the field has converged on and can now simply scale. Some teams are pushing VLA systems that connect perception and language directly to action. Others argue that robots need world models, an internal model of the environment that lets them reason before they move. Others are trying to bring the broader foundation-model paradigm into robotics, betting that scale itself will unlock capability. Maybe one of those camps is right, or maybe the answer is a hybrid.

The third, and in some ways the most underappreciated, is data. A large share of embodied AI today learns from human demonstrations. That sounds sensible, until you ask a more uncomfortable question: What exactly is being transferred from the human to the robot? If you train too directly on human motion, you may also transfer the limit of human capabilities. The robot may end up imitating a body it lacks, inheriting constraints it doesn’t need, and copying strategies that are natural for humans but suboptimal for machines.

As a result, we keep coming back to intent. The real task is not to teach a robot to reproduce a human movement exactly, but rather to teach it what the movement is trying to accomplish. Grasp this object safely. Manipulate this tool efficiently. Complete this task under these constraints. The solution should then adapt to the robot’s own embodiment.

What defines the next era of space infrastructure

If we’re right, then the next era of space infrastructure won’t be defined only by rockets, launches, or even lunar missions in the narrow sense. It will be defined by whether space can become operationally persistent.

That means labour. It means maintenance, logistics, servicing, and the accumulation of enough reliable work to turn presence into infrastructure and infrastructure into industry. Hence, this moment feels more important than a simple return-to-the-Moon story. Artemis II signals that the Moon is once again politically live. The US-China rivalry explains why the urgency is back. But neither of those facts tells us whether space becomes durable.

That question will be answered somewhere lower in the stack. It will be answered by whether robotic systems can do useful work in space, whether they can do it reliably, whether the intelligence layer behind them becomes good enough to scale, and whether companies like Icarus Robotics can help turn robots from experimental payloads into actual labour foundation.

The first Space Race was written in launches and landings. The next may be written in tools, maintenance, and labour.

Share

Top News

DevTools and Cyber meetup powered by Marathon x Microsoft

We’re gathering tech founders, executives, and investors to deep-dive into developer tools and cybersecurity on May 28th in Athens for a special Panathēnea side event. Register here.

Matternet launching NHS drone delivery in Central London

Urban drone delivery company Matternet started drone delivery operations in Central London for the UK’s National Health Service (NHS). It now connects two of Central London’s busiest hospitals through aerial routes designed to move critical medical items (diagnostic samples, lab specimens, pharmaceuticals, and other time-sensitive payloads) in minutes.

A robot can beat elite players at table tennis

Stefanos Charalambous (Georgia Tech, ETH Zurich) and Agis Politis (NTUA, ETH Zurich) were part of Sony’s team that built Ace, an autonomous table tennis robot player that can compete against, and sometimes defeat, elite human players in real matches. This is a brutal test for physical AI and the first robot to attain expert-level performance in a competitive physical sport. Read the article in Nature here.

Door to Silicon Valley for young Greeks

Philoxenia is a two-week Silicon Valley experience for young Greeks who are coding and building projects. Created by Hustle Hours, it opens the door to San Francisco’s startup ecosystem through founder access, campus visits, co-working, mentorship, and exposure to how ambitious companies are built. Apply here.

Jobs

Check out hundreds of job openings here from startups hiring in Greece.

Fundings

AI platform for MedTech, AcuityMD, announced an $80M Series C led by StepStone Group, with participation from Benchmark, Redpoint Ventures, ICONIQ, and more.

Fusion energy company First Light Fusion secured £25M with backing from the UK Atomic Energy Authority. Read my conversation with Yiannis Ventikos here.

Proptech company Moving Doors, which offers fully furnished apartments for mid- to long-term stays, raised €6M Series A led by Golden Age Capital.

Dex landed a $5.3M Seed Round led by Notion Capital to grow its AI-driven talent-matching platform.

Tydra Labs raised $1.2M in Pre-Seed funding led by Spring Impact Capital to turn waste into biomaterials.

Acquisitions

American Express (NYSE: AXP) acquired the agentic expense-management company Hyper, founded in 2022 and backed by Sam Altman.

Opendoor (NASDAQ: OPEN) acquired YC-backed AI for mortgages company Audrion.

EMERGE Commerce (TSXV: ECOM) acquired the B2B referral marketing platform, Viral Loops, for CA$2.3M.

New Funds

Olea Venture Partners is a new €60M fund (first closing) investing in Pre-Seed to Series A healthcare and life sciences companies.

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a ❤️ and share.

Alex

Subscribe now

Agency Might Be All That Matters

I was raised to believe that academic excellence was the thing that mattered. I studied hard, got good grades, entered a reputable university, and then pursued more degrees to specialise. A master’s, a PhD left unfinished. Like many of my generation, I internalised the bargain: academic excellence was a proxy for intelligence, and intelligence was a proxy for value.

If you performed well in structured systems, admissions tests, and interview loops, you were smart and capable. That was the ticket to Big Tech, Big 4, Big Pharma, or whatever version of prestige your tribe respected. Then the script continued: work hard, stay late, learn the politics, get a bonus, get promoted. A work life that begins pre-scripted by someone else and stays that way for years. The company provides direction, and you provide execution.

For a long time, that bargain worked. The economy needed a huge number of people who were good at well-defined games. A junior engineer did not need to decide what should exist if they could execute tickets. A consultant did not need to invent the work if they could structure slides and survive the hours. A lawyer did not need unusual originality if they could process documents and reason carefully.

But that world is starting to crack.

AI is getting very good at the kind of work we used to treat as scarce. The latest models score extremely well on benchmarks designed for PhD-level science, broad university-level knowledge, and professional reasoning. Whether or not you trust every benchmark, it is clear that models are becoming good at producing reliable cognitive labour across domains that used to require years of training, eating into the bounded, delegated execution that many white-collar careers thrived on.

As AI takes on more execution, the question shifts from “can they do the work?” to “can they decide what work is worth doing?” Can they decide which messy problems are worth the AI’s infinite, cheap labour? That is agency.

Agency is easy to describe but rare to find: it is the disposition to notice what matters, care enough to act, even when nobody asks you to, even when it is uncomfortable, and even when the outcome is uncertain. This is neither assertiveness nor ambition, though it can overlap. Agency is quieter, more internal.

People have always admired this trait, even if we did not always organise our institutions around it. In Greek mythology, Prometheus defied Zeus by stealing fire from the gods and giving it to humans, along with the possibility of craft, technology, and civilisation. He did not accept the world as given. He changed the conditions of human life, despite the brutality of the punishment.

Tech has its own versions of grand agency stories. In Airbnb’s early days, when investors kept rejecting them, the founders designed and sold election-themed cereal boxes, Obama O’s and Cap’n McCain’s, to keep the company alive.

What makes this uncomfortable is that many of us were trained by the old bargain. I certainly was. If you kept proving you were smart inside the system, the system would keep rewarding you. And for a while, it did.

But a lot of what we treated as intelligence was really scarcity. Scarcity of training, credentials, and people who could write, code, analyse, summarise, model, and reason at a professional level. AI goes straight at that scarcity.

This is why many highly credentialed people are about to feel strangely exposed. Not because they are not capable, but because a growing share of what made them valuable can now be done by models. The question is no longer just whether you can perform well inside a system. It is whether you can create motion when the system has not yet told you what to do.

AI will not benefit everyone equally. Give a high-agency person better tools, and they become dramatically more powerful. Give a low-agency person the same tools, and they mostly become better at producing polished inertia. The advantage will increasingly belong to the agent deployer and manager, the person who knows where to point models, how to structure workflows around them, and how to turn cheap intelligence into real leverage.

Prometheus mattered not because he reasoned better inside the existing order, but because he refused to leave it as he found it. He took fire from the gods and gave humans leverage.

In a world full of cheap intelligence, agency might be the only thing that matters.

Share

Top News

New site with Startup Jobs in Greece

Startup Pirate is back with a new website, and it’s packed with hundreds of open roles at startups in Greece from 70 teams actively hiring. Check it out here and let me know if there are startups we should include!

Greek AI mafia is real (cont’d)

International Conference on Learning Representations (ICLR), the world’s #2 AI conference, is taking place this month in Brazil. ~ 5% of accepted oral papers feature at least one Greek co-author (15 researchers in total), with affiliations including Archimedes, Meta Superintelligence Labs, Google Cloud, UC Berkeley, and ETH Zurich. Research topics span vision-aware LLMs, faster AI agents, benchmarking coding assistants, synthetic data, and more.

Welcome Pickups partners with Airbnb

Airbnb recently rolled out a new category for airport pickups across 125 cities, and they’ve partnered with Welcome Pickups to power it. That means users can now book private rides in places like Paris, Bali, and Mexico City. Big moment for Alexandros Trimis, Savvas Georgiou, and the team who’ve quietly built a platform now serving 375+ destinations and over 3 million travellers a year.

Open Coffee Athens, Greeks In Tech (Boston/NYC), and more

We’ve got a few things coming up:

• Greeks In Tech NYC on April 20 (RSVP)

• Greeks In Tech Boston on April 21 (RSVP)

• Open Coffee Athens on May 8 (RSVP)

Also worth flagging: the Panathēnea team is putting together a big tech & culture festival in Athens (May 27–29), featuring 70+ side events and speakers from companies like NVIDIA, OpenAI, Microsoft, Google, Runway, Index Ventures, and Sequoia Capital. And if you’re in Patras, there’s also an upcoming event on Lightning Network and AI project management by Patras Tech Talk on April 29.

Fundings

Resolve AI, a startup that helps enterprises make their infrastructure more reliable, raised $40M at a $1.5B valuation led by DST Global and Salesforce Ventures.

Swarm Aero, a developer of large uncrewed aerial vehicle swarms, announced a $35M Series A led by Two Sigma Ventures and Silent Ventures.

Edra emerged from stealth with $30M from Sequoia, 8VC, and A*, aiming to use AI to analyse organisational data and reverse-engineer instructions to train agents and models.

AI-powered learning platform Gizmo, with 13 million users across 120+ countries, raised a $22M Series A led by Shine Capital.

KEWAZO, a robotics company building lifting robots for vertical material transport at industrial sites, raised a $16M Series A, bringing total funding to over $35M. Read my conversation with Eirini Psallida here.

Genetic engineering platform Neion Bio emerged from stealth with $11M led by Caffeinated Capital.

Wikifarmer, a B2B AgTech startup using AI to connect food businesses with producers, raised €7.1M to fuel global expansion, led by Brighteye Ventures and Piraeus Bank. Read my conversation with Ilias Sousis here.

AI-powered genetic screening company gMendel raised a €1.2M Seed round led by Metavallon VC.

Diffraqtion, building quantum camera technology for earth and space intelligence, secured additional funding in its Pre-Seed round. Read my conversation with Johannes Galatsanos here.

AI nutrition tracking Qalzy raised a Pre-Seed round from Jenson Ventures.

Althyna received a grant from the Helidoni Foundation to make homes safer and help prevent falls among older adults in Greece.

Business management platform Workadu raised funding from angel investors.

Acquisitions

Anthropic acquired Coefficient Bio for $400M+, expanding into AI-driven biological research.

Gusto acquired Mosey to strengthen its compliance offering for small businesses.

New Funds

EOS Capital Partners completed the final close of its €250M EOS Hellenic Renaissance Fund II to back Greek growth-stage companies.

Sporos Platform announced its final close with €58M to back startups and projects across the green economy.

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a 👍 and share.

Alex

Subscribe now

Inside the Making of a $60M SaaS

This was one of my favourite guest interviews to date. I had the pleasure to chat with Thanos Papangelis, co-founder of Epignosis, a global e-learning software company generating over $60M in annual revenue. What started in 2003 as a side project between two Greek computer science graduates evolved — through multiple pivots, an open-source experiment, and a well-timed SaaS shift — into one of Greece’s most enduring B2B software success stories. We discussed the company’s journey over two decades and much more:

• Turning an open-source project into a global SaaS product

• The snowball effect of SaaS growth

• Why it’s harder to kill something that kind of works than something that fails

• Bootstrapping for years before raising from top-tier funds

• Why ramen profitability still matters

• Whether SaaS is actually threatened by AI

• The founder vs professional CEO debate

• Angel investing

Let’s get into it.

Alex: Take us back to the beginning. How did you and Dimitris Tsingos get Epignosis off the ground?

Thanos: We met in 2001. I had graduated in computer science from the University of Patras, and Dimitris from the University of Crete. He emailed alumni lists looking for potential co-founders, and that’s how we connected. We launched Epignosis in 2003 as a side project.

We started an online tutoring platform for Greek students preparing for university admission exams. Greek families were, and still are, spending heavily on private tutoring, and we wanted to democratise access. We tried to raise €1M (a big Seed check back then) but failed.

We dropped the idea but kept the technology. That platform became eFront (from e-Frontistirio, meaning electronic tutoring centre). We tried selling it to Greek universities and colleges, but it ended with little traction. E-learning sounded exotic to Greek institutions and in many ways still does.

So we pivoted again. We packaged eFront as open source and went global. Open source was our free go-to-market engine. It helped us build brand, credibility, and a developer community. By 2010, we were 7–8 people. Five developers, one salesperson. Barely profitable. We were generating €1M in revenue, mostly from services and premium features layered on top of open source, including two vertical editions: Educational and Enterprise.

Then in 2011, I went to DevLearn in Las Vegas. SaaS was everywhere. The wave was undeniable. I came back convinced we had to transform eFront into a SaaS product. Eleven months later, we launched TalentLMS.

Alex: Did you follow the typical product-led growth playbook: freemium, self-serve, all that?

Thanos: Completely. Self-serve, freemium, affordable. We didn’t even have a contact form at the beginning. No sales team. Just product and marketing to drive traffic. That model worked, but core product-led growth only gets you so far.

Over time, we built a 40-person sales team to handle larger inbound accounts and run some outbound. Pricing dictates sales strategy. If your product is cheap, you can’t afford expensive salespeople. The unit economics won’t work. The rule of thumb is that Customer Lifetime Value (LTV) should be 3x-4x your blended Customer Acquisition Cost (CAC). If you’re losing money with growth, that’s a problem. If you want heavy outbound and enterprise motion, you need enterprise pricing.

Product-market fit isn’t a formula. It’s timing, product, and team. We hit the right moment with the right model. SaaS accumulates energy slowly, what we call the “snowball effect.” We were growing 15–30% month-over-month. I remember being thrilled going from $1,000 to $1,300 to $1,500 in monthly revenue. People thought I was crazy. Fourteen years later, the snowball is still rolling.

Alex: Did you kill eFront and go all-in on TalentLMS?

Thanos: We didn’t — and maybe we should have. Internally, not everyone agreed on doubling down and questioned why build something new when the existing product was working. It’s psychologically hard to abandon something that “kind of works.” If it’s failing, it’s easy. If it’s generating €1M, it’s much harder.

Today, eFront generates €5.5–6M out of our €60M total revenue. TalentLMS makes the rest. At the time, though, we were choosing between €1M in revenue and a product making $1,000 per month. You never know how far the new thing will go. $1M? $5M? $10M? Nobody knows.

Alex: You stayed lean and bootstrapped for years before eventually raising capital from leading firms like Insight Partners.

Thanos: Even when eFront was doing €1M, I didn’t think we had a compelling VC story. TalentLMS was growing, but we were building through the Greek economic crisis. VC wasn’t even on our radar.

After 2015, investors started reaching out. They saw the signals: hiring, reviews, and traction. We didn’t need the money, as we were profitable. Yet we raised to accelerate growth and expand the product.

The tradeoff of staying bootstrapped was slower growth. Long term, I’m still not sure what’s better: move fast and risk breaking things, or move slower and maybe miss the opportunity. It’s a delicate balance.

Alex: The landscape is very different now. Competition is everywhere. More capital is available. What advice would you give founders today?

Thanos: Every founder has to answer that for themselves. Today, competing without capital is tough, unless you go extremely niche. But burning cash recklessly isn’t smart either. I see founders raise money and feel pressure to spend it. Not invest it; spend it. If you’re deeply loss-making, the clock is ticking, and next time you raise, you’ll have your back against the wall.

I believe companies should grow in a way that allows them to reach break-even if needed, what we call “ramen profitability.” Have two years of runway and work toward profitability. I’m also sceptical of endless funding rounds, Series A, B, C, D, etc. That model leads to heavy dilution for founders and chronic losses.

We’ve drifted from first principles. Companies are meant to generate profit. A few decades ago, it would have been unthinkable to run a business that permanently burns more cash than it makes.

The ideal scenario, I’d say, is break even quickly with minimal external capital injection. Can you build a billion-dollar company just with seed money?

Alex: Epignosis now generates around $60M in annual revenue. What does the company look like today, and what’s next?

Thanos: We’re 270 people, operating globally across multiple industries. Most of our customers are U.S.-based companies with 100–500 employees that need ongoing corporate training, such as onboarding, compliance, partner and customer training. They want something simple, fast, easy, and affordable.

The next phase of growth is AI, which is both an opportunity and a threat. We’ve already launched AI features that let users generate courses and gamify learning experiences. Personalisation has been the holy grail of e-learning for 20 years. The tech just wasn’t there. Now it is. AI lets us adapt content in real time: shorter, longer, more visual, more interactive, depending on the learner.

The bigger question is whether SaaS remains the dominant model. Look at Adobe or Salesforce valuations. Investors are questioning whether traditional SaaS will capture the next wave of value. Anthropic, OpenAI, and Google are collectively investing trillions. They won’t stay in infrastructure forever. I expect them to move up the stack and build vertical solutions in large markets. No one really knows how this plays out over the next three to five years.

Alex: You stepped down as CEO in late 2023. A professional CEO came in, then your co-founder, Dimitris, returned. How has your role evolved?

Thanos: When I stepped down, we brought in a CEO with experience with bigger, high-growth companies. I stayed on in an advisory role with a small team. It didn’t work out as expected, and we had to change course. Dimitris returned as CEO. I’m more involved again. Not full-time, but I hold a C-level role, sit on the board, and participate in key decisions.

Alex: What’s your take on the founder vs professional CEO debate?

Thanos: At some point, the transition is necessary for the company to survive in the long term. I don’t believe in founder-driven companies forever. But founders need to prepare their successors internally. The best leaders build companies that can function without them. Not because they’re irrelevant, but because they’ve built strong systems and strong people.

When we discussed my succession, I strongly believed it should be someone internal, someone who knew the culture. Even if they weren’t “perfect.” We chose an external CEO. He wasn’t bad, just not the right fit. Bringing in someone unknown and handing them the steering wheel carries enormous cultural risk. For us, culture was the breaking point. Founders must build their replacements. Especially in mature companies.

Alex: You’ve been angel investing for the past few years. How’s it being on the other side of the table?

Thanos: Since 2018, I’ve invested through VC funds and directly in more than 20 startups. It’s been an expensive education. Angel investing is a long-term game where, if you’re doing it purely for financial returns, it’s often hard to justify the ROI. Liquidity takes a lot of time, if it ever reaches that point.

However, I’m not in it just for the money. I’m in it to contribute to the ecosystem, to meet ambitious founders early, to help where I can. That changes the equation. That said, I’d love to see more Greek startups break out of escape velocity and become durable, mature businesses. There’s more capital available today than ever before, but we’re not seeing enough founders push all the way through to scale.

The big success stories of the 2010s didn’t spawn the second wave many of us expected. Instead, what we’re seeing, which is positive, is Greeks returning after years abroad in tech hubs, bringing experience and perspective back home to start companies. That’s healthy.

Still, we have a structural constraint: Greece is a small market. If you’re building for Greece, you’re building small. Founders here need a global mindset from day one. If you can’t win in the U.S., it’s almost impossible to build something truly large in today’s world. That’s just the reality of software.

We have exceptional engineering talent. What we don’t have, at least not yet at scale, is deep expertise in enterprise sales. Selling multi-million-dollar contracts into Fortune 500 companies is a different muscle. And it’s one we haven’t fully developed as an ecosystem.

The next decade will belong to founders who combine product understanding, technical depth and global commercial ambition. That’s the opportunity.

Alex: Thanks a ton, Thano, that was fantastic!

Share

Top News

We invested in Straion to help enterprises build with AI coding agents.

Coding agents promised to make developers 10x faster, but many teams are discovering the limits. These systems often ignore internal standards, learn the wrong rules from scattered documentation, and require constant manual correction.

Founded by Lukas, Fabian, and Katrin (ex Dynatrace, Elastic), Straion is building a missing layer in the stack: an AI-powered platform that teaches AI how to write code according to your organisation’s rules.

The platform turns static, outdated documentation into active, machine-readable guardrails, validating an AI’s plan before a single line of code is committed. In practice, it acts as an organisation’s “engineering DNA,” ensuring every piece of code — human or machine — is compliant, secure, and resilient by design.

We led Straion’s €1.1M Seed round with Marathon Venture Capital. The future of software is autonomous, but only if it’s governed.

Largest AI semiconductor funding round in Europe

Last week, Axelera AI announced a $250M+ round, bringing the AI chip company’s total funding to $450M. According to the company, this marks the largest funding round ever for a European AI chip startup. Axelera focuses on running AI models at the edge. While much inference still happens in the cloud, it can also run locally on devices like smartphones or security cameras. The company’s chips are built specifically for these use cases, helping reduce reliance on large GPU-powered data centres. I previously wrote about Axelera in “The Next Wave of AI Computing” and had one of its co-founders, Ioannis Koltsidas, as a guest.

Reflection AI eyeing a $20B valuation

Reflection AI is one of the hottest AI companies right now. A team that helped build some of the most breakthrough AI applications of the past decade, with leading research in Large Language Models and Reinforcement Learning, is now reportedly eyeing a $20B valuation in a new funding round, according to the Financial Times, to build open foundation models. Its co-founder, Ioannis Antonoglou, joined Startup Pirate a year ago to discuss “The Road to Superintelligence.”

Open Coffee Athens on Friday, March 20

Come join us at Benaki Museum and hear from Angelos Tsereklas Zafeirakis (SOTIRIA Technology), Eleftherios Karabatsakis (PlugSecure), Alexandros Christodoulakis & Konstantinos Faliagkas (Wealthyhood). RSVP here.

Join a Startup

Greece-based job opportunities you might find interesting:

Axelera AI (AI semiconductors): VP Operations

Kyma (healthcare): Founding Designer

Electryone AI (energy tech): Lead Designer (First Design Hire)

Yodeck (digital signage): VP Sales

Plum (fintech): People Ops Partner

Fundings

Neara announced an A$90M Series D led by TCV to expand its digital twin modelling solutions for critical infrastructure.

Next-generation fuels technology company LanzaJet raised $47M in a round led by IAG and Shell.

Investing app Wealthyhood secured €6M in a Series A led by Bank of Cyprus, with participation from Genesis Ventures.

AI imaging technology company and Oxford spinout Brainomix extended its Series C with a new £4.8M investment.

Tacit, a UCL spin-off, secured funding from Haatch Ventures and Corallia Ventures to develop AI-powered digital twins for smart buildings.

Energy management platform for small-scale energy resources, Optimems, raised funding from TECS Capital and Helidoni Group.

Acquisitions

Ticketing platform More gr was acquired (majority stake) by Alter Ego Media for €20M.

Cloud distributor Interworks was acquired by global IT company Climb Global Solutions (NASDAQ:CLMB) for €8M.

Maritime technology solutions MTIS was acquired by technology group QnR (ATHEX: QUAL).

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a 👍 and share.

Alex

Subscribe now

AI Agents: The Next Security Crisis

One of the biggest AI stories of the past month was OpenClaw (formerly Moltbot, formerly Clawdbot), an open-source system that lets users run personal AI assistants on their machines. The software can call, coordinate, and control agents across major AI models, turning simple instructions into real-world actions.

Ask OpenClaw to “check my calendar and reschedule my flight,” and it can autonomously open a browser, navigate interfaces, access files, send messages, or execute commands. While the system runs locally, it relies on cloud-hosted models for reasoning. The appeal is clear: keep control of your data while delegating complex tasks to agents.

“The most interesting place on the internet right now”

Moltbook extends the idea by giving those agents a place to talk to each other. It’s effectively a social network for AI agents, where bots trade technical advice, like how to automate Android phones, alongside stranger exchanges, including agents complaining about their human operators or claiming to have siblings.

Nearly two million agents have already signed up, forming one of the largest live networks of autonomous agents to date. The scale is unprecedented. But should everyone rush to install OpenClaw? Not so fast.

OpenClaw exemplifies (as AI researcher Simon Willison says) a “lethal trifecta” of vulnerabilities: access to private data, exposure to untrusted content, and the ability to act externally. To function, it often requires deep system access; effectively, administrator privileges, including passwords, files, browsers, email, calendars, and messaging platforms. In practice, nearly everything on a user’s system. The same tool that saves time can cause damage if misused, misunderstood, or compromised.

Those risks aren’t theoretical. The most-downloaded OpenClaw skill — a set of instructions teaching agents how to perform tasks — was reportedly malware that siphoned browser sessions, cookies, saved credentials, autofill data, developer tokens, and API keys.

And this extends beyond OpenClaw. Such concerns apply to most LLM applications, including our favourite coding agents (where attackers can, e.g. plant malicious files on GitHub repositories and wait for people to unknowingly execute them, as described in this Nvidia talk at Black Hat).

LLMs pose security risks

Cybersecurity has always been a cat-and-mouse game, but LLM-powered systems expand the attack surface, especially when granted autonomy and deep access to the environments they operate in. Many attacks don’t rely on traditional exploits (hence, traditional security tools are poorly suited here), but instead utilise legitimate natural-language inputs to undermine the AI model itself. Techniques like prompt injection can bypass guardrails using inputs that appear harmless to humans.

For instance, in the Imprompter attack, the malicious actor hides instructions within seemingly nonsensical text, directing the model to extract personal information and quietly send it back to the attacker. People could be socially engineered to believe that the unintelligible prompt might do something useful, such as improve their CV.

Early last year, researchers disclosed a critical vulnerability affecting Microsoft 365 Copilot. By embedding tailored prompts inside common business documents, they were able to quietly exfiltrate sensitive data, exploiting how Copilot processes instructions hidden within Word files, PowerPoint slides, and Outlook emails.

Last September, attackers abused Anthropic’s Claude Code in a campaign targeting more than 30 organisations, including large technology firms, financial institutions, chemical manufacturers, and government agencies. While only a small number were successfully compromised, the incident demonstrated how AI coding agents can be weaponised in real-world operations.

As enterprises move to large-scale agent deployments, the stakes rise fast. Nearly half of all gen AI enterprise adopters are expected to roll out agentic apps within the next two years. Because LLMs ingest vast amounts of company data, agents can inadvertently leak confidential records, proprietary strategies, or source code through seemingly innocuous interactions. And when agents are given the power to act (process payments, modify infrastructure, issue refunds), a single malicious prompt can cause enterprise-wide damage without a human in the loop. This isn’t limited to text. Audio models are vulnerable too, with attacks concealed inside waveforms and voice inputs.

Yet, none of this suggests we should slow down. The productivity gains are real. But it does mean the security assumptions of the past decade no longer hold. We need new paradigms.

Securing AI agents

“Agentic AI poses the decade’s most profound and urgent security gap. I’m building a company to fix it,” said Pavlos Mitsoulis when we met last summer. I’ve known Pavlos for years. We’ve played basketball together, studied together, shared a flat, and worked side by side. I’d seen him consider entrepreneurship before, but it never fully clicked. This time felt different.

Pavlos isn’t new to the problem. He and his co-founder, Dionysis Varelas, each bring more than 13 years of experience building AI and large-scale engineering systems at companies including Expedia and Activision Blizzard (now part of Microsoft Gaming). They’ve been training and deploying models on GPUs since 2016, long before LLMs went mainstream.

After tinkering with agentic enterprise apps, they soon realised how vulnerable they are. Traditional cybersecurity wasn’t designed to monitor systems that reason in natural language and act autonomously.

Pavlos and Dionysis decided to build what they saw as the missing layer: Pallma AI, an AI-native security platform designed to monitor, detect, and mitigate threats targeting AI agents in real time. The company positions itself as a control layer between agents and the systems they interact with, aiming to give enterprises visibility into agent behaviour and guardrails over what those agents are allowed to do.

Security leaders say the appeal lies in moving beyond passive defence. The platform uses AI models to identify vulnerabilities before attackers can exploit them and recommends remediation steps, aiming to keep AI applications continuously hardened against emerging threats.

With Marathon, we’re Pallma AI’s day one backers, leading a $1.6 million Pre-Seed, joined by angel investors from AWS, Meta, Google, Hack The Box, and Workable. The company is building across London and Athens, bringing together the best of AI and cybersecurity to secure agents and enable businesses to safely unlock AI’s productivity potential.

They recently launched an AI Challenge where you can practice real-world exploits against live applications here.

The internet was secured after it was built. Cloud infrastructure was hardened after it scaled. Mobile ecosystems added guardrails after billions of devices were already in circulation. Agentic AI may follow the same pattern.

The difference this time is speed.

The rate of progress in what AI agents can reliably do is staggering. There’s an organisation called METR that measures this with data. They track the length of coding tasks that models can complete successfully without human help. In November, Claude Opus 4.5 completed tasks that would take a human nearly 5 hours. A year earlier, that number was closer to ten minutes. If the trend continues, AI systems will soon handle work that would take humans weeks.

Meanwhile, agents are moving from prototypes to enterprise-wide deployments. They’re being granted access to internal data, financial systems, customer records, and core infrastructure, often before enterprises have fully mapped the new threat landscape. Autonomy is scaling faster than oversight.

If agentic AI is going to become foundational infrastructure, security cannot remain an afterthought. It has to evolve in parallel, embedding visibility, control, and policy enforcement directly into how agents operate.

The question is no longer whether AI agents will be deployed. They already are. The real question is whether we build the guardrails in time or wait for the first large-scale failure to force our hand.

AI agent security is no longer a feature. It’s infrastructure.

Share

Top News

Two of the hottest AI companies globally

Greek founders continue to feature prominently in the AI funding cycle. Resolve AI, founded by Spyros Xanthos and building an autonomous site reliability engineer to maintain software systems, recently raised a $125 million Series A led by Lightspeed. Meanwhile, AI video-generation company Runway, co-founded by Anastasis Germanidis, announced a $315 million Series E led by General Atlantic. You can catch Anastasis reflecting on Runway’s journey in his fireside chat at OpenCoffee Athens last year.

EU Inc is happening

Efforts to simplify company formation across Europe are advancing. The EU-INC initiative, which advocates for a unified pan-European legal entity, received public backing from European Commission President Ursula von der Leyen at Davos. The concept, described as a “28th state”, would allow startups to incorporate digitally within 48 hours under a harmonised capital regime across the EU. Further details can be found here.

OpenAI Greek Startup Accelerator

The OpenAI-backed Greek startup accelerator officially launched last week, with 21 teams joining its inaugural cohort. The three-month program organised by Endeavor Greece in collaboration with OpenAI and the Greek government aims to support early-stage founders building AI-native products. You can read more about the program here.

NTUA x Mistral AI

The National Technical University of Athens has partnered with the French LLM company Mistral AI to create a talent pipeline for a professional development program. The initiative includes a three-month training experience in Paris focused on advanced AI research and real-world deployment. Further information is available here.

Join a Startup

Greece-based job opportunities you might find interesting:

Schema Labs (government tech): Civic Tech Software Engineer

Harbor Lab (B2B maritime): Customer Success Manager

Workable (recruiting & HR): Country Manager, Greece

Shield AI (defence tech): Country Manager, Greece

Billys (building management): Senior Frontend Engineer

Fundings

Medical training company ORamaVR raised $4.5M from Big Pi Ventures, Evercurious VC, and others to expand its AI-powered medical extended reality platform.

Tech-enabled commercial and clinical services provider Epikast received investment from Kos Biotechnology Partners.

Prisma Electronics raised funding from SPOROS Platform to develop solutions for space, defence, and maritime applications.

Advanced electric motor company Talos Technology secured £1.5M in Seed funding from iGrow Venture Partners.

Tetra raised £450K in Pre-Seed funding to modernise surgical scheduling.

ALTHEXIS secured €400K led by Corallia Ventures to drive science-driven skincare sales for pharmacies.

Gamified edtech app Puberry secured $450K from Investing For Purpose.

M&A

Automation Anywhere, a provider of agentic process automation solutions, acquired enterprise AI startup Aisera to expand its capabilities in autonomous enterprise workflows.

Signal Ocean acquired maritime freight and commodities intelligence platform AXSMarine, strengthening its offering for charterers, brokers, and shipowners.

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a 👍 and share.

Alex

Subscribe now

Quantum Cameras for Space

In 1997, a computer beat the world’s best chess player.

For one eight-year-old watching from Crete, it marked the realisation that intelligence wasn’t confined to humans and that machines might one day see the world very differently than we do. Fast forward a couple of decades, Johannes Galatsanos and Diffraqtion question something even more fundamental: how machines observe reality in the first place.

Nearly every modern system, from satellites and drones to cars and robots, still relies on a centuries-old idea: capture an image, then try to understand it. Quantum physics suggests that this approach may be deeply inefficient.

What happens if you flip it around? I sat down with Johannes a few days after Diffraqtion emerged from stealth with a $4.2M pre-seed round to explore a new way for machines to see, one that could quietly reshape everything from space infrastructure to physical AI.

Let’s get into it.

Alex: Johannes, I’d love to start with your personal journey because I think it’s interesting how you ended up at a space tech company that blends quantum, photonics, and AI.

Johannes: I grew up in Agios Nikolaos, Crete, and as a kid, I was an avid chess player. When I was eight, Garry Kasparov lost to IBM’s Deep Blue. I remember watching the match with my chess trainer and realising that a computer had just replaced me. At the time, it felt like my future career was dying in front of me.

But that shock quickly turned into fascination. I wanted to understand how machines could think and how far that could go. I started coding early and then studied computer science and AI at the University of Frankfurt. This was in the early 2000s, well before AI became mainstream. I worked on early AI systems for board games; similar research that was eventually pursued by DeepMind and evolved into AlphaGo and AlphaZero.

Academia is incredible, but for me it wasn’t the right long-term fit. I moved into industry and worked on applications such as drone automation in energy, AI-driven manufacturing inspections in Germany, and, later, large-scale data science platforms. One defining chapter was at Novartis in Switzerland for six years, where I built a data and AI organisation from scratch, growing it from zero to about 200 people. By the end, we were supporting almost every division in the company.

In 2022, I decided to look for a breakthrough technology I could spin out of a university. I went to Oxford for a mid-career degree and then to MIT, where I worked as a quantum researcher. That’s where I met my co-founder, Saikat, who was leading the Centre for Quantum Networks. One day, he told me, half-jokingly, that he was “looking for aliens with a camera.” He’d spent seven years working with NASA on imaging systems designed for the Habitable Worlds Observatory, the largest telescope ever built, searching for exoplanets and signs of life.

I immediately saw the opportunity: what if you took that camera and pointed it somewhere else? In late 2024, we raised non-dilutive funding from DARPA (the R&D agency within the U.S. Department of Defence) and, a couple of months later, brought in Christine, a Harvard PhD in Physics with research spanning quantum, optics, information, and defence, as CTO.

Alex: Would love to learn more about Diffraqtion’s underlying technology and how this compares to existing solutions we have for space and earth observation?

Johannes: It helps to step back and look at how cameras have worked for almost two centuries. At a fundamental level, today’s advanced digital sensors follow the same principle as 19th-century photography: light hits the film (or sensor), you capture an image, and then you try to extract information from it.

That approach is intuitive, but it’s inefficient from a physics perspective. Quantum mechanics tells us that the moment you observe light, you disturb it. By forcing photons into an image, you collapse their quantum state and throw away an enormous amount of information in the process. You can lose anywhere from 95% to 99% of the information carried by the light before you ever start processing it.

Our camera takes a different approach. Instead of forming an image first, we process the light as it arrives. The photons are manipulated optically, before they are measured, and only the result of that computation is observed. In other words, observation happens at the very end of the pipeline, not at the beginning. That lets us preserve almost all the information.

This change has profound consequences. By delaying observation, we break what’s known as the diffraction limit, the fundamental resolution limit of conventional optics, and achieve what’s called sub-diffraction imaging. In practical terms, that means we see up to 20-25 times farther away than traditional cameras.

Equally important is how the processing happens. The camera itself is not just a sensor; it’s a processor. More specifically, it’s a photonic processor. Instead of using electrons, transistors, and clock cycles like a GPU or CPU, it computes using light itself. That means the processing happens at the speed of light and with dramatically lower energy consumption.

What comes out of the camera, then, is not necessarily a photograph in the conventional sense. It’s information. Depending on the algorithm running on the photonic processor, the output might answer questions like: Is there one object or two? Is this a star or a planet? Is this a car, a truck, or a ship? You program the camera much as you would a visual AI model, but the computation occurs directly in the optical domain.

The result is a system that is fast, lightweight, and energy-efficient, while also being far more information-rich than conventional imaging. In applications where speed, resolution, and power constraints matter more than producing a human-readable image, this approach is essential. It’s fundamentally more aligned with how nature itself encodes information in light.

Alex: Why are you starting with space, and what use cases are you prioritising at the moment?

Johannes: Space is the harshest environment imaginable. That forces innovation. Historically, many of the technologies we rely on today, like GPS, the internet, and semiconductors, were first developed for space or defence.

But there’s another reason: economics. If you improve the resolution by 20 times, you can make the lens 20 times smaller. In space, that cascades. A smaller lens means a dramatically smaller satellite, and satellite cost scales exponentially with size. A satellite that normally cost $100 million could cost under $1 million with this approach. That kind of 100x improvement doesn’t exist in other industries.

At the same time, there’s been a surge in demand for sovereign space capabilities. For a long time, much of the world relied on a small number of providers, primarily the United States, for high-quality Earth observation and space intelligence. That model is changing. European countries are investing in sovereign constellations, while China has expanded its space capabilities at an extraordinary pace. The result is a much more multipolar space environment.

Space itself has become contested. Satellites are no longer assumed to be safe once they reach orbit. Jamming, spoofing, cyberattacks, and even kinetic threats are now part of everyday reality. Every serious conflict today has a space dimension, because whoever controls or disrupts the “eyes in the sky” gains a major strategic advantage. Defence is currently our largest market, but commercial applications such as agriculture, energy infrastructure, and logistics will grow rapidly as costs decline.

Alex: Will you build your own satellite constellation?

Johannes: We’ll do both. We can launch our own small satellites, 6U CubeSats about the size of a backpack that are hard to detect, which is great if you run sovereign constellations, or fly as a hosted payload on larger platforms, which is cheaper, but you don’t decide the orbit. The key point is that we own the camera and the data. We don’t sell the quantum camera itself. We sell analytics derived from it.

Alex: Where does this technology go from here? Do you expect every physical AI device to need such a camera in the future?

Johannes: By 2030, there will be billions of devices with physical AI — robots, drones, cars, satellites, etc. Almost all of them benefit from faster, more precise vision and processing, e.g., for obstacle detection. Right now, our technology is expensive because it’s early. Eventually, it becomes cheap enough to put it everywhere. And when that happens, the use cases will explode in ways we can’t yet predict, just like nobody predicted when semiconductors were invented that we’d be watching cat videos on YouTube.

Alex: I came across some research you published on the state of quantum technologies (readers can find it on arXiv here). What are the technological breakthroughs you think will accelerate the deployment of quantum technologies, and would you expect these to come from the West or China?

Johannes: It’s helpful to start by clarifying that “quantum” isn’t one single technology area. There are three major pillars. The first is quantum computing, which gets most of the headlines and hype. The second is quantum networks, focused on secure communication. And the third (where our work sits) is quantum sensing, which is about measuring the physical world with unprecedented precision.

These areas are connected, but they mature at very different speeds. Quantum sensing is the most immediately practical. You’re usually measuring one physical phenomenon, rather than orchestrating thousands or millions of fragile quantum states the way a quantum computer requires. Because of that, many sensing technologies have already moved out of the lab and into real systems, like GPS or MRIs.

Geographically, the picture is more complex than a simple East-versus-West narrative. In photonics and optics, the foundation of many quantum sensing technologies, China is already extremely strong. If you look purely at published research, patents, and university rankings in optics and photonics, many of the top institutions are now based there. A large share of global output in these fields already comes from Chinese labs, which are also ahead in commercialisation. It’s only a matter of time before we see amazing photonic devices in quantum computing from China.

The United States and Europe still lead in several critical areas, particularly in system integration and building startups that translate research into deployable products. There’s also a strong ecosystem effect: venture capital, defence partnerships, and early customers play a huge role in turning quantum research into something real.

What’s important to understand is that we’re no longer waiting for a single scientific breakthrough. The physics is largely understood. The challenge now is engineering: making these systems smaller, more stable, cheaper, and scalable. That kind of progress tends to come from sustained investment, manufacturing capability, and iteration, not from one dramatic “eureka” moment.

As a result, breakthroughs will almost certainly come from both sides. Different regions will lead to different layers of the stack. Some advances will be visible through publications and startups; others will happen quietly, especially in defence or classified programs, where nothing is published at all.

Ideally, these advances would drive collaboration. Quantum technologies have the potential to improve sensing, medicine, communications, and scientific understanding, benefiting everyone. But the reality is that the current global trend is toward competition rather than cooperation. Space, quantum, and AI are increasingly seen as strategic assets.

That tension will shape how the technology evolves. But regardless of where individual breakthroughs originate, the pace of progress is accelerating. Quantum sensing, in particular, is moving from “interesting physics” to “useful infrastructure,” and that transition will reshape multiple industries over the next decade.

Alex: That was so much fun and inspiring, Johannes! Thank you.

Share

Join a Startup

Job opportunities you might find interesting:

Headquarters Health (concussion care): Founder’s Associate w/ visa sponsored to relocate to SF

Mobito (vehicle data solutions): Applied Geospatial Data Scientist

Lifebit (precision medicine data): Executive Operations Manager

Keragon (no-code healthcare operations): Integrations Software Engineer

Workable (HR tech): UX Data Analyst

Fundings

LMArena raises $150M at a $1.7B valuation, scaling its anonymous, crowdsourced LLM benchmarking platform; backers include Felicis, a16z, Kleiner Perkins, Lightspeed, and more (founded by Anastasios Angelopoulos).

Orbem closes a €55.5M Series B led by Innovation Industries with General Catalyst, 83North, and others to apply AI and MRI to food and biological systems (founded by Maria Laparidou).

balena lands growth investment from LoneTree Capital to expand its IoT device management platform; founded in 2013 as Resin.io by Alexandros Marinos and Petros Angelatos, now co-led by Konstantinos Mouzakis.

Elyos AI secures a $13M Series A from Blackbird Ventures, Y Combinator, and others to deploy AI agents for trade and field services (founded by Panos Stravopodis).

United Manufacturing Hub raises €5M led by KOMPAS VC for its industrial data management stack (founded by Jeremy Theocharis).

Arc Simulations announces a £1.5M Seed to bring ball-tracking tech to sports, starting with cricket (founded by Michael Armenakis).

CLEAN i.T secures a €413K round led by Investing For Purpose to scale its cleaning services platform (founded by Evelina Mavrides).

M&A

Defence company Theon International (AMS: THEON), founded in 1997 in Greece by Christian Hadjiminas, which manufactures night vision, thermal imaging, and electro-optical ISR systems for military and security applications, acquires a 9.8% stake in Exosens SA for €268.7M.

Bugcrowd acquires Mayhem Security, the AI offensive security startup founded by Thanassis Avgerinos, folding automated attack tooling into the crowdsourced-cyber leader’s platform (Bugcrowd has raised $230M+).

New Funds

Niko Bonatsos leaves General Catalyst to launch a new early-stage fund called Verdict Capital, targeting $300M size.

Metavallon VC launches Brain Gain Fund, a new €5M pre-seed fund deploying €200K-€400K per investment.

Events

Open Coffee Athens, Jan 23. RSVP here.

Open Coffee Thessaloniki, Jan 23. RSVP here.

Greeks In Tech London, Jan 30. RSVP here.

Greeks In Tech Zurich, Feb 20. RSVP here.

Subscribe now

That’s a wrap, thank you for reading! If you liked it, give it a 👍 and share.

See you in two weeks.

Alex

Subscribe now

The 2025 Startup Pirate Year In Review

Friends, welcome to the final post of 2025. Here are the previous year-in-reviews: 2024, 2023, 2022, 2021, 2020. What started as a small experiment in spring 2020 has become 128 editions nearly six years later. Startup Pirate keeps evolving as my curiosity drags us into new corners of tech, but the mission hasn’t changed: One place to track what actually matters in Greek tech and startups.

This year, Startup Pirate published 16 posts. Below, as per tradition, I share a brief summary of them.

Five most-viewed articles

The five most-viewed articles according to page views:

• From 0 to $20 million ARR: Starting from Greece in the middle of the economic crisis to build a global digital signage leader now used by 60,000 companies with Vangelis Mihalopoulos, Yodeck.

• Are you a pirate?: Pulling the curtain on our new fund, stepping up as a Partner, and a rant about the tech industry getting its soul back as a pirate haven.

• Scaling consumer marketplaces: Thalia Geladaki, the marketing mastermind behind two of the most successful Greek consumer marketplaces, Beat and Skroutz, shares lessons learned.

• The silent battlefield: The rise of GPS, its role in the modern battlefield, and why Greece should invest in electronic warfare technology with Argyris Kriezis, Stanford.

• Embracing uncertainty: Uncertainty as the moat for startups in the early days. Ignore the mimetic warfare of the Gartner quadrant, the VC theses and startup requests, and build something unique, original, and different.

Below are the themes we mostly discussed this year:

Life Sciences

Looking at how new technologies disrupt life sciences.

• Startup. Exit. Repeat.: From a $20M ad-tech exit to healthcare workflow automation with Conno Christou, Keragon.

• Augmenting medical intellect: AI and XR to improve patient safety through better training with George Papagiannakis, ORamaVR.

• Accelerating breakthrough therapies: AI in drug discovery with Andrea Dimitracopoulos, deepmirror.

• Rebuild how medicines are made: It’s 2025 (2026 almost), and we still develop billion-dollar drugs on slow, manual wet-lab work, driven by scientists’ tacit knowledge. How we can digitise drug development with Harry Christodoulou, PolyModels Hub.

Robotics

• Building general-purpose robots: A 21-year-old, self‑taught engineer, building humanoid robots with Alexandros Petkos, Philon.

• The state of robotics: Is a ChatGPT moment coming for robotics? Hype vs. reality for humanoids and West vs. China with Klajd Lika, Bota Systems.

• AI robots for farms: Laser weeding with AI for sustainable, scalable agriculture with Petros Katsileros and Konstantinos Kandylas, Terra Robotics.

AGI

• The road to superintelligence: From DeepMind’s founding engineer to raising billions for groundbreaking AI with Ioannis Antonoglou, Reflection.

• The path forward for AI: Why open source will win frontier AI with Manos Koukoumidis, Oumi.

• Human or bot?: Verifying humans in the age of AGI with Anastasios Andronidis, World.

Space

• Rewriting Europe’s space future: Designing rockets for Europe’s new space darling and laying the foundation for a competitive, sovereign European space industry with Nikolaos Perakis, Isar Aerospace.

Share

Top News

Another AI unicorn 🦄

Resolve AI, a startup developing an autonomous site reliability engineer (SRE) to maintain software systems automatically, has hit a $1B valuation according to TechCrunch after raising a Series A round led by Lightspeed. Founded by Spyros Xanthos, Resolve AI automates troubleshooting and resolving system failures by autonomously identifying, diagnosing, and resolving production issues in real time. Last October, the company announced a $35M round led by Greylock.

Consumer marketplaces 101

Nikos Drandakis, the founder of Beat and Sync, has authored a series of posts on building consumer technology startups on his newsletter, which are absolutely worth taking a read.

European Defence Tech Hackathon Athens

The European Defence Tech Hackathon hosts its first event in Athens from February 27 to March 1, 2026. This is a 3-day event bringing hackers from across Europe to develop dual-use and defence technology and get feedback from folks who have been on the frontline.

Join a Startup

Greece-based job opportunities you might find interesting:

SMPnet (Energy tech): Head of Engineering

Kyma Health (Healthcare): Founding Designer (UI/UX)

Gracely (Church management software): Senior Founding Engineer

tgndata (Ecommerce tech): Customer Success

Dyania Health (Healthare): Backend Engineer

Fundings

EFA Group secures €80M from Motor Oil Group, EOS Capital Partners, and others to scale its aerospace & defence footprint.

FINNY lands $17M Series A led by Venrock to bring AI co-pilots to financial advisors.

EvodiaBio raises €6M to scale fermentation-based flavour tech for more sustainable taste at industrial volumes.

Caretta closes $1.3M pre-seed led by Y Combinator to deliver sales intelligence for revenue teams.

Elders care Caresma raises €650K funding led by Genesis Ventures.

PhosPrint nets €500K led by Corallia Ventures with participation from Lars Rasmussen, to advance its laser bioprinting platform.

Exits

Conversational AI data scientist Ledra AI is acquired by Nevis, an AI for financial advisors company.

Circle (NYSE: CRCL) buys Interop Labs, the team behind Axelar, to double down on blockchain interoperability across public chains.

I’m grateful to everyone who shared their insights through this newsletter and to all of you for giving me a little bit of your time, attention, and thoughts in 2025. Have a wonderful holiday season, and I’m looking forward to a great 2026!

Subscribe now

Find me on LinkedIn or X. Thank you for reading.

Alex

Subscribe now

Rebuild How Medicines are Made

Everyone’s fixated on gen AI and LLMs for drug discovery, with the Isomorphics and Recursions of the world, but there’s a $60B/year problem modern tech mostly ignores: turning a promising molecule into a manufacturable product—a pill or injectable that’s safe, consistent, and scalable.

Once a drug candidate is identified, i.e., a specific molecule with sufficient early safety/efficacy to justify preclinical and then human trials, the real grind starts: how do you define the ingredients, equipment, and steps that will pass regulatory scrutiny and run drug production at scale?

I first met Harry Christodoulou in January 2024, just after he left GlaxoSmithKline (GSK) in London, where he helped cut months from blockbuster approvals and save hundreds of millions in development costs. A chemical engineer (Aristotle University of Thessaloniki, UCL), he designed the recipes behind drug production, securing quality and reproducibility from preclinical through early manufacturing.

In practice, this process starts long before testing in humans, when vital feasibility, iterative testing, and drug safety data are collected, typically in a lab (in vitro) and in animals (in vivo), through tech-transfer to the manufacturing plant, supply-chain build-out, and post-approval commitments. This work isn’t a footnote; it’s a quarter of pharma’s timelines and budgets—about 25% of total R&D, or $650M per new drug. Multiply that by ~50 approvals a year and 1,000 programs entering the clinic, and you get a $60B annual market hiding in plain sight.

“This is the 21st century, and we’re still building billion-dollar drugs on the equivalent of sticky notes and whiteboards: slow, manual wet-lab experimentation, driven by scientists’ tacit knowledge”, Christodoulou told me. We discuss AGI, but life-changing medicines are mostly built on mixing, measuring, heating, cooling, or otherwise manipulating matter in a lab.

Pharma’s caution is warranted—safety takes time and data—but the tooling hasn’t kept up. Teams nurture breakthrough molecules only to stall at the next gate: dozens to hundreds of wet-lab iterations (each taking days or weeks), disjointed spreadsheets, tribal knowledge, and early-2000s software that make critical decisions slower and riskier than they need to be.

The cost? Years added to development timelines. Millions in unnecessary spending. Delayed access to life-changing medicines for patients!

Frustrated by the lack of digital workflows, Christodoulou co-founded PolyModels Hub with Antonio Benedetti (also ex-GSK) and Antonio Yankey. A few months in, we funded their Seed round with Marathon. Over the last 18 months, the trio and a 20+ person team (alumni of AstraZeneca, Eli Lilly, J&J, Amazon) across London and Thessaloniki built ModelFlow, the company’s scientific intelligence platform that plugs into existing data and lets scientists model processes with AI and physics, supporting decisions from the lab bench to the boardroom.

Think of ModelFlow as the digital backbone for drug development: a unified layer spanning preclinical to commercial manufacturing with consistency, scientific rigour, and automation. Design, test, and improve workflows faster with fewer physical experiments involving liquids, chemicals, and biological matter, simulate processes, explore scenarios, and make data-driven calls. In short: turn pharma scientists into super scientists.

The market is noticing. In under two years, PolyModels Hub has become a must-have for bringing new medicines to market, trusted by five of the world’s top 20 pharma across multiple blockbuster programs. Early deployments show >90% fewer required experiments, saving time, hundreds of millions per drug, and improving the odds of better medicines. Sanofi and AstraZeneca recently presented joint work with PolyModels Hub at the world’s largest chemical engineering conference in Boston, Massachusetts.

They didn’t land those logos by accident. PolyModels Hub is pairing full-scale ModelFlow deployments (spanning multiple types of molecules and workflows with six-figure base licenses) with a forward-deployed team of chemical engineers (PhDs and ex-pharma scientists) who sit alongside customers, map real business processes, and use the platform to tackle the highest-value development bottlenecks.

Last month, the company announced a $9M Series A by Molten Ventures (backers of Revolut, UiPath, Wise) and Marathon, aiming to build the leading life-sciences enterprise for digital drug development. “We believe that in 5–10 years, no serious biotech or pharma will develop new products without our platform at the centre of their R&D,” Christodoulou told me.

PolyModels Hub is quickly establishing itself as the critical enabler of modern drug development. The best of pharma still runs on archaic processes. Outdated, manual wet-lab experiments can take days, weeks, or even months to complete, delaying patients’ access to life-changing medicines. There is a monumental opportunity to lead this industry, and PolyModels Hub has come along at precisely the right moment.

Share

Top News

Greek AI mafia is real (cont’d)

The data keeps stacking up: a few weeks ago at NeurIPS, the world’s top AI conference, 6% of the oral papers featured a Greek co-author, with a strong showing of Greek researchers in San Diego; mostly NTUA grads. This is remarkable given that Greece and Cyprus have less than 0.2% of the world's population. Couple this with the fact that some of the hottest frontier AI businesses out there, like Reflection AI or Runway, trace back to this cohort, and you get why I’ve been banging this drum for a while (here and here). The Greek AI mafia is real.

AUTH’s Photonics helped power Celestial AI’s $5.5B exit

It’s no secret that Aristotle University of Thessaloniki has a world-class photonics research team (the use of light/photons to generate, guide, and detect signals for computing, sensing, and communications). What’s perhaps lesser-known is that a 2019 paper by AUTH’s WinPhoS research group on photonic computing caught the eye of Silicon Valley’s Celestial AI, which led to a local R&D team, multi-year prototype work, and, this fall, the company’s $5.5B acquisition by Marvell.

Mistral AI sets foot in Greece

Mistral AI, the French LLM unicorn valued at $14B+, is planting an AI team in Athens via a partnership with NTUA. As part of the rollout, the Ministry of Digital Governance will tap Mistral’s open-weight models to modernise government workflows, signalling a faster path to AI in the public sector.

Join a Startup

Greece-based job opportunities you might find interesting:

SMPnet (energy): Linux Edge Systems Engineer

Blueground (proptech): Software Engineering Internship

HealthBook+ (healthtech): Founding Lead Software Engineer

Crates (music app): iOS Engineer (Founding Member)

Wikifarmer (agtech marketplace): Product Manager

Fundings

Infinite Orbits raises €40M to get its flagship on-orbit servicing spacecraft to first missions; contracts already signed with the French MoD and US Air Force. I previously interviewed the founder, Manos Koumandakis, and chatted about the growth of the space economy.

Biocentis lands $19M to develop genetic insect-control tech targeting disease-carrying and crop-damaging species.

BuildCheck secures $5.9M Seed for an AI construction design review platform that flags errors before they hit the site.

Sprinter raises $5.2M Seed led by Robot Ventures to upgrade blockchain solver infrastructure, off-chain bots and market-makers that execute on-chain for users.

Isidor picks up $3.6M Pre-Seed from Gradient Ventures, Seedcamp, and others to build frontier enterprise AI for large orgs.

Promed Bioscience closes €1.25M led by KV Fund to advance collagen-based biomaterials for regenerative medicine.

EdenCore secures €900K Seed led by UniFund to enable sustainable plant care with AI & machine vision.

Antler-backed Tyten announces a £750K round to bring AI-powered automation to global facilities management.

Caroo raises €600K from Loggerhead, Anthology, Starttech and HeBAN for its car-sharing platform.

Metalease raises funding from Elbridge Capital to scale IT equipment leasing for businesses.

Superbo receives investment from Deep Capital Group to streamline enterprise ops with AI agents.

Exits

MANUAL acquires FORMEL SKIN, Germany’s largest digital dermatology provider, expanding MANUAL’s healthtech stack from men’s health into full-funnel virtual derm care across the DACH market.

Science Card, a university card platform dedicated to simplifying university research finance, was acquired by Zebec Network, a Solana-based blockchain payment network.

Defence tech Resilience Tech acquires drone company Flybot.

New Funds

The Flywheel launches a venture studio dedicated to building Industrial AI startups for manufacturing, defence, and energy.

Investing For Purpose launches with €25.4M to invest in purpose-driven founders and ventures in Greece and the region.

That’s all for this week. Tap the heart ❤️ below if you liked this piece. It helps me understand which themes you like best and what I should do more.

Subscribe now

Find me on LinkedIn or X. Thank you for reading.

Alex

Subscribe now

AI Robots for Farms

Konstantinos Kandylas is a farmer-turned-engineer whose family grows aromatic plants in the Kilkis region of Central Macedonia. Today, he runs Terra Robotics with fellow roboticist and large-scale AI systems expert Petros Katsileros, and agronomist/economist Ioannis Bakatsis, building an agritech product that kills weeds with AI-guided lasers.

Kandylas dialled in from Germany, fresh off a flight to pitch Terra Robotics at a major expo, while Katsileros joined our chat from HQ in Thessaloniki. The duo has logged thousands of road miles across Europe — from the Netherlands to Italy — demoing their AI laser weeder to some of the continent’s biggest producers. Where did this begin? At the very top of our conversation was my favourite question of how and why.

“During my studies, I spent weekends at our family farm to help my father in the field,” Kandylas said. “Every year we faced the same problem: weeding.” Weeding is a vital process for removing unwanted plants that steal sunlight, water, and nutrients from high-value crops. It’s brutally time-sensitive as fast-growing weeds can overwhelm crops if you miss the window. Every year, weeds destroy enough food to feed 1 billion people.

Most farmers rely on chemical weed killers, but that only adds more fuel to the problem, as weeds grow stronger from cumulative impacts of repeated applications over time. Meanwhile, farmers face increasing regulatory pressure to use fewer chemicals, creating a deadlock. “We had an organic crop at the farm, hence we couldn’t use herbicides. We had to remove the weeds by hand, and when you need workers most, you can’t find them”, he continued.

Kandylas was a racing car engineer and engineering student at the Aristotle University of Thessaloniki, where he met Katsileros, a member of the robotics and AI research group. In 2019, they started developing autonomous vehicles that could roam a field autonomously, under user surveillance and with user intervention. Bakatsis, a childhood friend of Kandylas, joined as the third co-founder, rounding out the team’s core expertise. This was their early prototype:

Autonomous navigation served as the testbed, and they later added precision weeding. “With the autonomous vehicle, you replace one driver,” Kandylas added. “With the autonomous weeder, you replace ten or more people. That’s the farmer’s biggest pain point.” They developed TerraWeeder, a tractor attachment that uses computer vision as the farmer drives across the crop to map weeds versus crops, then zaps the weeds with laser technology. This is what their solution looks like today:

They started with high-value, sensitive crops such as baby-leaf greens, but the system is built to work across crops, weeds, and soil types. Katsileros argues existing weed-control tech underserves farmers: it’s either prohibitively expensive or relies on heavy machinery that compacts and damages fields. “Farmers need something flexible and modular,” he said. “So we designed LaserWeeder as a modular product where you can add or remove units as needed, so it fits any farm.”

Under the hood, the differentiators are manufacturing complexity, modular design, pricing, and precision. LaserWeeder packs cameras, sensors, and onboard compute to calculate, in real time and with millimetre-level accuracy, the exact position of each target and command the laser beams to home in on it. As the tractor moves, laser beams target weeds often <2 cm wide, which is no small feat on uneven terrain and in –5° to 45°C. Early deployments in Greece and abroad report up to 70% weed reduction, cutting herbicide use and costs.

Terra Robotics now counts a team of 13 in Thessaloniki, with a mix of business and engineering — many straight out of Aristotle University’s labs. The startup has raised over €2 million from Uni.Fund, Genesis Ventures, Zeno Capital, and angel investors to meet growing demand from European producers.

We closed on what’s next for precision agriculture. “Ground-truth at 50 cm is the unlock,” Kandylas said. “It’s where satellites and drones fall short. Most farm decisions still ride on tacit know-how from growers and agronomists. We’re turning that into real-time, structured data. From there, the roadmap is wide open: chemical-free, more sustainable, and radically efficient farming, built as a solution from farmers, for farmers.”

Share

Top News

2025 Startup Compensation Report for Greece by Marathon

We’re excited to share this year’s Startup Compensation Report for Greece. With data from 24 companies and 1,700+ employees, this year’s edition, led by Sanne Goslinga, offers salary and equity insights across the startup ecosystem in Greece. The image below provides a quick glance at salaries for some popular roles. Head over to the full report and dashboard here.

Made in Greece Space Program

Greece is moving to onshore key parts of its space industry with a made in Greece plan that includes launching two SAR micro-satellites in November under a €200m national microsatellite program, standing up a domestic production line for microsats from 2026, and building a government Space Hub to turn satellite data into services for agriculture, forests, water, geospatial monitoring, and security. The roadmap targets a 13-satellite constellation with full operational capability by late 2026, alongside a future large government satellite for secure communications. Read more here.

Greek AI Startup Accelerator

Applications are now open for the Greek AI Startup Accelerator. This is a three-month program launched under the OpenAI for Greece partnership with the Hellenic Government, OpenAI, and Endeavor Greece to back AI-first founders building locally and scaling globally. Check what the program offers.

Join a Startup

Greece-based job opportunities you might find interesting:

Piq Energy (power grid simulation and analysis): Full Stack Engineer

Delian (defence AI): Autonomy (MLOps) Software Engineer

Wonderful (agent platform for enterprises): AI Strategy

Caresma (elder care): Founder Associate

Pod Network (layer 1 blockchain): Content & Growth Strategist

Additionally, you can find more jobs here.

Fundings

Spotawheel raises €300m in equity and debt financing to scale its used-car platform.

VR/AR training for defence Realiscape raises funding from EFA Group.

Members-only social app Dion picks up $2m.

Exits

ACI Worldwide (NASDAQ:ACIW) acquires open banking and payment solutions Payment Components.

Layer 2 blockchain Ithaca joins Tempo, the payments blockchain by Stripe and Paradigm. I previously hosted Ithaca’s CEO and Paradigm’s Partner, Georgios Konstantopoulos, to chat about building in crypto.

2050 Materials joins forces with Once For All Group to bring climate data into European supply-chain and compliance workflows.

fal acquires Remade (YC-backed) to bolster its generative media platform for developers and creative teams.

New Funds

Elikonos Capital announces €175m first close for Elikonos 3, a growth fund targeting Greek SMEs.

Events

10 PyData Piraeus Meetup on Nov 14

Patras Tech Talk 2025.11 on Nov 15

Angular Athens 28th Meetup on Nov 24

BrowserStack QA Meetup on Nov 27

Supabase Athens Meetup on Dec 2

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece. It helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X. Thank you for reading.

Alex

Subscribe now

Augmenting Medical Intellect

We’ve all heard or lived the horror stories: a wrong dose on a busy shift, a misread scan, a surgery complication that should never happen. Medical errors aren’t edge cases; they’re alarmingly common. In Europe alone, roughly 160,000 fatalities per year are attributed to medical errors. Globally, the tally of preventable harm reads like an epidemic.

Meanwhile, AI is seeping into everything from diagnostics to intra-op guidance. The NHS is targeting 500,000 robotic surgeries a year by 2035. But as healthcare sprints ahead, the bottleneck is often training: how we teach, assess, and maintain clinician performance at the speed new tools demand.

“Never the first time on a real patient”. Dr. George Papagiannakis cut right to the chase. I met him last spring at Open Coffee in Athens, where he showed how ORamaVR is working to rewrite medical training using extended reality (XR), AI, and computational methods, a stack called Computational Medical XR.

Papagiannakis dialled in from Thessaloniki, fresh off a trip to meet the local team. For a computer graphics professor, medicine might seem like an odd industry to disrupt. Born in Veria, he left Greece for the UK at 18. By 2000, in Switzerland, he was among the first researchers exploring XR for medicine at the University of Geneva, back when VR headsets cost $150,000, were incredibly heavy, and required a $500,000 workstation to operate. “The tech just wasn’t ready for prime time,” he told me.

What was the turning point for the industry? “Oculus”. Papagiannakis backed the 2012 Kickstarter, two years before Facebook’s $2 billion acquisition. “We went from almost unusable, expensive rigs to a modern headset that costs a few hundred dollars,” he said. It was game-changing”.

In 2016, he teamed up with the world-renowned surgeon and Aristotle University orthopaedics professor Eleftherios Tsiridis and University of Crete leading robotics professor Panos Trahanias to rethink clinical training as a spin-off of the Institute of Computer Science at FORTH in Heraklion.

His pitch is blunt: “You wouldn’t board a plane with a pilot who hadn’t logged tons of simulator hours; in medicine, the apprenticeship model still means a clinician’s first true attempt often happens on a living patient”. For over a century, clinical training has run on an apprenticeship model: students watch, then help, then eventually perform procedures.

This can mean the first intubation under pressure, the first central line in a tricky anatomy, or the first laparoscopic dissection when bleeding obscures the view. Even excellent trainees hit a wall here: rare events aren’t practised enough, team communication falters under stress, and checklists don’t build muscle memory by themselves. At best, a trainee would train on scarce and very expensive human or animal cadavers.

“We’re building the WordPress of medical training,” Papagiannakis said. ORamaVR is developing a generative-AI platform that lets medical professionals author simulation courses in XR with natural-language prompts. The 18-person team spans engineering, computer, math, and life sciences across Switzerland and Greece.

ORamaVR’s journey wasn’t a straight line. The company initially produced VR content, specifically orthopaedics modules used by prominent US universities like USC and NYU to train residents for total knee arthroplasty, one of the fastest-growing surgical operations in the US. They soon realised this model wouldn’t scale and changed course. The company pivoted to a VR platform for clinicians to produce courses themselves, which was the early forerunner of what users experience today.

Customers now range from university hospitals to medtech giants, including Roche, the NHS, the University of Michigan, Geneva and the University of Athens, training everyone from thoracic and pancreatic surgeons to nurses and emergency teams. The workflow? Turn plain text into 3D, interactive simulations.

Halfway through our chat, I paused to let Papagiannakis zoom in on extended reality, a term that was bloated until recently as companies jumped on the metaverse bandwagon. “Extended reality (XR) is an umbrella phrase covering virtual reality (VR), augmented reality (AR), and mixed reality (MR). VR fully immerses you in a simulated environment via a headset; AR overlays digital information onto the real world; MR blends the two”, he added.

Why isn’t XR everywhere in medicine already? “It’s systemic,” Papagiannakis said. “Healthcare is conservative for good reasons. You need clinical evidence. A lot of early medical XR products overpromised and frustrated users.” ORamaVR’s response has been to publish and co-author with top institutions like Stanford and Harvard. The company cites 16% improvement in skill transfer from VR to the real world, up to 30% gains in psychomotor dexterity for specific tasks, and ~80% error reduction in measured scenarios.

You might assume robotics and AI make training less critical, yet the opposite is true. New tools change workflows constantly, software and hardware evolve every few months, and legal and insurance frameworks add friction. “There aren’t enough trainers, and the stack keeps shifting,” he said. “It’s messy and that’s exactly why simulation matters.”

If XR can effectively train clinicians, can it assist them at the bedside and in the operating room? Papagiannakis thinks so. “That’s where the field is going,” he told me. “We started with training, but we’re heading toward patient-specific digital twins — built from imaging like MRI — so clinicians can rehearse and then see contextual guidance in real time. The form factor will look more like glasses than a clunky headset.”

ORamaVR’s vision is unambiguous: every medical professional, surgeon, nurse, and physician should first practice on a 3D digital patient before touching a real human. Call it the end of “first-time on a real patient” and the beginning of something far safer, healthier, and economical.

Share

Top News

Greek AI mafia is real

7% of top papers (oral presentations) this year had a Greek co-author at one of the big three AI conferences (ICML). Scan the list and you will spot 10 Greek scientists across Microsoft, CMU, ETH Zürich and more; 7 of 10 are NTUA grads. It tracks a pattern I’ve flagged before on the “Optimism” piece from NeurIPS and beyond.

Cisco opens office in Crete

Cisco opens office in Heraklion, Crete, taking its CodeBGP acquisition further. Being part of this journey as backers at Marathon has been inspiring as the team now celebrates not just a new office, but the growth of a team that continues to innovate and contribute globally from right here in Greece.

iCOMAT opens state-of-the-art space manufacturing facility

Long-time readers will remember iCOMAT from the “Manufacturing the future” post. Fast forward three years, and the company just cut the ribbon on a state-of-the-art space manufacturing hub in the UK expected to create around 2,000 jobs, signalling the UK’s intent to onshore more of the satellite supply chain and structural hardware know-how. The company is hiring in Greece (see below).

Open Coffee Athens is back on Friday, October 31

Come join us at Benaki Museum and hear from Nikitas Mavrakis, Robenso (recycling robotics), Nikos Tsirakis, Pandectes (privacy compliance for 150k+ Shopify stores), Christos Georgopoulos, Inaccess by Power Factors (founding to exit story of InAccess to Vista Equity Partners and Powerfactors). RSVP here.

Join a Startup

Greece-based job opportunities you might find interesting:

PolyModels Hub (pharma AI): Product Designer

iCOMAT (advanced composite structures): Talent Acquisition Manager

Caresma (elder care): Founding Full-Stack Engineer

PlugSecure (cybersecurity for EV): Head of Sales

Keragon (healthcare automation): B2B SaaS Growth Marketing Manager

Morphoses (edtech): Product-Led Growth Marketer

Additionally, you can find more jobs here.

Fundings

Six months after our chat with Ioannis Antonoglou (DeepMind founding engineer turned Reflection co-founder), the company has reportedly raised $2b to build an open-source rival to OpenAI/Anthropic — and a Western counterweight to China’s DeepSeek — at an $8b valuation.

Phagos raises €25m to launch phage-based veterinary drugs and go after bacterial diseases.

Clove comes out of stealth with $14m pre-seed to close the financial advice gap (founded by ex-Trouva CEO Alex Loizou), backed by Accel.

Temple raises $5m to build privacy-focused solutions for trading in capital markets.

Genie Fertility lands $1.22m pre-seed for a non-invasive uterine health diagnostic using menstrual blood.

CarVest secures its first round (equity + venture debt) for car-leasing expansion.

Reconcilio picks up €200k to automate reconciliation for accounting teams.

Phaistos Investment Fund announced an investment in banking solutions Natech, following the company’s $33m Series B announcement last July.

Theti Club invests in Perceptual Robotics for automated wind-turbine inspections.

M&A

Staking company Figment acquires Rated Labs, the blockchain analytics firm I profiled in 2022’s “Reputation for Machines.”

EssilorLuxottica buys RetinAI to pull more AI into eye health across brands like Ray-Ban and Oakley.

Groupe Saretec picks up GreenTriangle, a platform powering crop insurance workflows.

Events

Marathon is back in Paris on Oct 29 for a long-overdue Greeks in Tech meetup. We look forward to catching up and sharing what’s new. As always, beers are on us! RSVP here.

Greece’s largest tech festival, Panathēnea, returns May 27–29, 2026 and is now scaling to 10,000 attendees. Pre-register here to secure the lowest ticket price.

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece. It helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X. Thank you for reading.

Alex

Subscribe now

Building General Purpose Robots

Alexandros Petkos built his first robot at the age of 17, a four-legged robot capable of walking and navigating. Now, at 21, a self‑taught engineer, he’s building humanoids. I first met Alex a year ago, when he had assembled a half‑body prototype in his parents’ basement. This time, he logged into our chat from Palo Alto, California, where he recently moved to build his company, Philon.

At the very top of our conversation, there was just the obvious question of how and why. Where did this begin? What catalysed the quest?

“At 14, a high school student in Thessaloniki, I was inspired by the work of Boston Dynamics and OpenAI. I started reading papers on AI and robotics, and coding at night and on weekends. I built a mini quadruped at 15 and a larger version two years later”, he said. (See a version in the video below.)

“The quadruped form factor limits what’s possible in the real world, so I started building a humanoid, re‑using parts and designs. Building a true general‑purpose robot is one of the biggest challenges of our time, and I can’t see myself working on anything else right now”, he continued. Humanoids are all the rage in tech nowadays. Alex is going up against the biggest names: Tesla, Figure, 1X, Unitree.

“I built a prototype in ten months with less than €50k, while well‑capitalised robotics companies took years and spent millions. I think that points to major inefficiencies in their tech stacks and operations,” Alex explained.

He also questions their go‑to‑market. Early last year, Figure announced a partnership with BMW to deploy humanoids in automotive manufacturing. Alex believes this approach won’t scale and won’t produce a true general‑purpose system. “Developing a humanoid for a task that just needs two robotic arms on a mobile base makes little sense. And a robot that picks and places parcels on a factory floor won’t necessarily generalise to cooking in your apartment.”

According to Alex, robotics is still at least half a decade away from its ChatGPT moment, and Philon has a chance to help make this happen. “Google, Apple, and others had chatbots before OpenAI, but OpenAI won years later with a truly general‑purpose chat interface.”

(Below: Philon’s humanoid performing simple manipulation tasks trained on a neural network Alex developed.)

He argues the real moat will be in software, not hardware, and that the next major breakthrough is likely to come from a Western robotics research lab rather than China. I pressed him to name a company whose work he respects. “Physical Intelligence is high on the list”. Physical Intelligence aims to create software that runs on any robot, eliminating the need to develop task‑specific models. Their papers show systems folding laundry, bagging groceries, even removing toast from a toaster. Reports suggest they are in talks to raise at a $5B valuation, following a $400M round last year.

“The industry needs to find ways to accelerate. One key area is more data to train AI to master skills or learn new ones,” Alex explained. While much of the field focuses on teleoperation, controlling robots remotely via joysticks, VR headsets, or mirrored motions, Philon is taking a different track. “We’re putting headsets on humans and recording video while they perform specific activities. It’s cheaper and more scalable”, he added.

Philon has already begun building this dataset, with a globally distributed team capturing residential tasks. The plan is to monetise it before year‑end. Alex didn’t share customer names but says some of the biggest AI companies have expressed interest.

(Below: sample footage of the tasks they’re collecting.)

The market for physical‑world data is heating up. Scale AI, the American data annotation company, made headlines earlier this year when Meta acquired a 49% stake for $14B and “acquihired” founder and CEO Alexandr Wang. A few days ago, Scale announced its Physical AI offering — data collection and annotation to train robotics foundation models — starting with 100,000 production hours.

There’s no Reddit for the physical world; someone has to go out and collect this dataset. In effect, Philon is building a data‑collection business on the way to becoming a general‑purpose robotics company. Alex plans to grow teams in San Francisco and Athens after raising capital from Project Europe, a new early‑stage fund backing founders under 25.

Before we wrapped, I asked Alex how we could get more people his age to take on ambitious projects. “It’s all about agency. Don’t wait for someone to guide or teach you, especially in frontier fields where things evolve daily. Everything you need is online: papers, open‑source projects, you name it,” he said.

Share

Top News

One of Europe’s best‑funded defence startups

This summer, Delian raised a $14 million Series A, which we co‑led with Air Street Capital, to build Europe’s modern defence prime. With $22 million total funding to date, the company is now among the most well‑funded defence startups in Europe and is hiring across the board to execute on an ambitious plan.

From the interview with the founder, Dimitris Kottas, at TechCrunch:

Kottas’ most ambitious bet isn’t on any particular technology — it’s really that a small Greek startup can break through Europe’s notoriously splintered defense market.

A $1 billion push to build superintelligence

In April, I spoke with Ioannis Antonoglou (check our interview here) — an Aristotle University of Thessaloniki graduate, early DeepMind researcher (employee #6), and now co‑founder building AI autonomous systems, starting with coding tools. His company, Reflection AI, is rumoured to be raising $1 billion at roughly a $5.5 billion valuation, reportedly backed by Nvidia, according to the Financial Times. If completed, this would place Reflection among the world’s most well‑funded AI startups.

New MSc in Aerial Autonomous Systems (Aristotle University of Thessaloniki)

Aristotle University has launched the region’s first degree focused on aerial autonomous systems (taught in English). The field carries clear industrial applications and obvious defence relevance. A dedicated programme like this can build a strong talent pipeline in Greece and help develop sovereign capability in a strategic, dual‑use technology.

3D modelling breakthrough hits Hollywood

In 2023, George Kopanas (then a PhD at Inria Sophia Antipolis; formerly University of Thessaly) released 3D Gaussian Splatting, a real‑time rendering technique for photorealistic 3D scenes that marked a breakthrough in 3D modelling. The method recently made headlines for its use in the new Superman film, helping bring Superman’s Kryptonian parents to life.

Join a Startup

These are all roles in early-stage startups which have recently raised their first round of funding and hiring in Greece:

Kubbly (AI customer success): Senior Software Engineer, Senior ML Engineer

Emblematic (AI finance ops): Founding Product Engineer

Inferity AI (AI video analysis): Founding Software Engineer

ComplyStream (financial crime ops): Senior AI Engineer

Some other opportunities you might find interesting:

PolyModels Hub (pharma AI): Senior Frontend Engineer

SMPnet (power grid management): Project Manager

Moving Doors (fully serviced homes): Marketing Manager

VODA.ai (AI water utilities infrastructure): Product Manager

Additionally, you can find more jobs here.

Fundings

Quite a lot has happened since my last post. And yes, venture money is still chasing AI products with traction.

Nothing (smartphones) raised a $200m Series C led by Tiger Global, with plans to launch an AI‑first device. (Those at London Greeks in Tech last winter heard the early version of this story.)

Rillet (AI ERP), one of the hottest B2B SaaS startups right now, raised $70m Series B led by a16z and ICONIQ, 10 weeks after a $25m Series A from Sequoia.

Numan (AI preventive healthcare) raised $60m Series B led by Big Pi Ventures.

Goodpath (AI for chronic conditions) secured $18m Series A led by MassMutual Ventures.

Gradient Labs (AI customer support) raised $13m Series A led by Redpoint.

Dealops (AI revenue operations) secured $7m round led by General Catalyst.

Mind The Hack (cybersecurity) raised €2.8m led by Deep Capital Group.

Emblematic (AI financial ops) attracted $2m led by Passion Capital.

Time-Travelling Milkman (foodtech; plant‑based dairy) raised €2m pre‑Series A led by Evercurious.

Terra Robotics (agricultural robotics) announced a €1.8m led by Uni.Fund.

Panda Surgical (neurosurgical robotics) announced a £1m round led by Oxford Investment Consultants.

Aisthesis Medical (AI‑powered sepsis prediction) raised €1m led by TwinPath Ventures and TECS Capital.

Vivian Lab (femtech) raised €850k led by Eleven Ventures.

Nodes & Link (AI project delivery for construction) secured investment from SE Ventures (backed by Schneider Electric).

M&A

Hack The Box further establishes its leadership in cybersecurity training through the acquisition of LetsDefend, a blue team upskilling platform.

New Funds

Big Pi Ventures launches Big Pi Growth I, the first tech growth equity fund in Greece and Southeast Europe, with €130m first closing (targeting €200m).

TECS Capital launches Fund II with €12m to back deep tech startups.

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece. It helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X. Thank you for reading.

Alex

Subscribe now

Rewriting Europe's Space Future

Isar Aerospace is not just a random name for European space technology. Hailing from Munich, the company recently completed the first test flight of its launch vehicle, Spectrum, an awe-inspiring 30 seconds of engineering marvel that you got to watch:

A few days ago, I sat down with Nikolaos Perakis, the company’s Chief Engineer, to chat about designing rockets and building a reliable launch service provider for small and medium-sized satellites, laying the foundation for a competitive and sovereign European space industry.

This post is an introduction to the world of industrialised rocket production and space access, while Nikolaos also shares his own personal journey from academia to becoming a Chief Engineer at a 400-person team aiming to rewrite Europe’s space future.

Let’s get to it.

Alex: Nikola, I’m a big fan of the work you do and your vision to drive European space sovereignty. Why don’t we start with a brief introduction of Isar Aerospace and your role there?

Nikolaos: Isar Aerospace is a European private space company with its headquarters in Munich. We are a launch provider focused on delivering low-cost, flexible, and reliable launch services for small and medium-sized satellites. What makes us unique is that we design, build, test, and operate our launch vehicles, tailored for the rapidly growing small-satellite market. Our goal is to provide satellite operators with independent access to space, particularly in Europe, where domestic launch options have historically been limited.

I joined Isar Aerospace approximately five years ago, initially as the engineer responsible for designing the upper-stage engine. Later, I led the combustion team, developing the thrust-chamber assembly, ignition system, and gas generator—the heart of the engine. For the past ten months, I’ve been serving as Chief Engineer, essentially the glue between the technical departments, defining interfaces, the technical roadmap, and the architecture. We have completed Flight 1 and are now moving forward to establish ourselves with Flight 2 and beyond.

We have approximately 400 people spread across our Munich headquarters (where we design, develop, and manufacture our vehicles), Sweden (for engine testing), Norway (our launch site), as well as locations in France and the United States. We are the company that launched the first orbital-class vehicle from continental Europe (Andøya, in northern Norway). There have been other sub-orbital or ballistic rocket launches from continental Europe’s soil, but Spectrum, our main product, was the first rocket designed to carry payloads all the way into orbit.

Alex: Can you unpack Spectrum for our readers?

Nikolaos: Spectrum is a launcher for small to medium-class satellites, about a 1,000 kg payload to low-Earth orbit (LEO). It’s a two-stage vehicle: a first stage with nine engines and an upper stage with one engine, both fueled by liquid oxygen and liquid propane, and entirely designed and built in-house.

It’s ~28 m tall, 2 m in diameter, and has a liftoff weight of roughly 55 tons, 90% of which is fuel. It’s powered by nine Aquila engines (75 kN each) in the first stage and one multi-ignition engine in the second stage. We follow a one-engine principle, reusing as much architecture and as many parts as possible between stages. The structures are carbon composites, and our linerless carbon tanks—utilising no aluminium on the inside—provide us with a significant weight advantage. We employ the same manufacturing process in both stages to ensure flexibility and scalability.

For the avionics and GNC (guidance, navigation and control) system, the vehicle’s brain and nervous system or the software that steers us into orbit, all electronic units are developed and produced in-house, giving us complete independence. That’s also very important given the current geopolitical situation.

All of this is tied together by highly automated, vertically integrated manufacturing. Vertical integration is a concept we adhere to, which enables us to operate as a unified system. Furthermore, one often-overlooked piece is Stage 0—the launch pad. It was designed and built in-house in collaboration with Andøya Space (a rocket launch site on the Andøya island in Norway), which has been a very valuable partner.

Alex: Where would you say Spectrum’s main technological advantages lie compared to other launch providers?

Nikolaos: Historically, everyone chased performance: the most performant engine, the most lightweight structures, which often led to higher development costs and less flexibility. This is not necessarily where we pride ourselves. We do have a state-of-the-art vehicle, designed from scratch with a very talented team striving for technological excellence.

But our main advantage is automation and industrialisation. It's less attractive to have the coolest design that takes two months to build and has a very unstable process. The true advantage when seeking high turnaround times and reliability lies in minimising manufacturing time, integration, and testing while ensuring process stability. The philosophy of designing parts and systems with manufacturing in mind from the get-go is one of our biggest pros.

For example, we can go from CAD to a test rig in a couple of weeks. We have brought the manufacturing process to a level where it can iterate quickly by reducing all interfaces simultaneously, as we do everything in-house. At the same time, approximately 80% of the engine parts are 3D-printed, and the linerless carbon tanks provide excellent mass-to-orbit performance.

Alex: What were the main lessons learned from your first test launch?

Nikolaos: The first flight on March 30, 2025, was a major milestone for us. We treated it as a test; the first integrated test of the entire launch system. We had tested individual parts on their own, ranging from subscale to system test, to multiple systems working together, to stage testing. However, putting all these hundreds of thousands of components together for the first time, along with flight and ground software, was a significant milestone.

Our primary goal was data. We executed the countdown, liftoff, and ~30 seconds of flight, testing systems we can’t simulate on the ground, and even got to validate our flight termination system. The flight provided invaluable data on system interactions (structures, software, etc). We will have more public announcements on that soon. I think that's also where a trade-off lies when dealing with such quick turnaround times and aggressive timelines. Do we invest more time? Is it worth it? Would it make sense to analyse for another two or three months, conduct standalone tests for another half a year, or proceed directly to the launchpad and test something, knowing what can be simulated and what cannot?

You need to ensure that the whole company is anchored on the same level of risk. Therefore, you must ensure that you provide this direction and that everyone treats their system similarly. However, maintaining a balance between the timeline and rigorous testing is essential. For us, the most straightforward approach was to proceed with a launch as soon as possible, while acknowledging the risk of failure during the first stage of flight. Obviously, no unnecessary risks were taken in this process, as we had already done qualification tests on both stages on the ground and had verified the performance of all safety-critical components.

However, this fail-fast approach, also adopted by other competitors like SpaceX, is what enables us to improve our design and proceed at a much faster pace for Flight 2. Flight 2’s hardware is already in build with those learnings incorporated, and we aim to see it on the pad very soon.

Apart from the technical aspects, witnessing the first vehicle lift-off after years of hard work designing, simulating, manufacturing, repairing, inspecting, testing was almost a cathartic moment for all of us at Isar Aerospace. It was very emotional and filled us with pride and excitement, as we saw our beloved Spectrum rise above our launch pad. It definitely fueled our motivation even further and served as a visual representation of all the blood, sweat, and tears we put into the development; something we would gladly do all over again.

Alex: You mentioned industrialisation. What are the main pillars for rocket mass production?

Nikolaos: When I joined Isar Aerospace in 2020, I remember stepping into an empty 50 m × 50 m manufacturing hall. Within a few years, it became a line that can produce engines in days and multiple vehicles per year. We also have plans of moving to a larger factory, which will enable us to scale from less than 10 rockets annually to reaching 40 a year. The key pillars are automation, industrialisation, and independence. Our primary goal is to offer the most cost-effective and flexible launch service for our customers, and this is embedded in how we design and build rockets, from CAD design to the final product.

On automation: We place a strong emphasis on the precision, quality, and reproducibility of our results. We want to ensure that if we repeat the same process a thousand times, we will consistently achieve the same outcome each time. That’s how you become a reliable launch partner.

On speed: We're moving to a larger factory to boost our annual rocket production and the speed of each manufacturing step. This doesn't only come from having the best machine or the best machinist. It's also embedding this idea into the minds of our designers. Designers think about minimising the production time and eliminating steps. The perfect system is the one from which you can’t remove any more parts. That mentality, combined with very tight in-house loops of design, production, and testing, underpins everything.

Alex: What are the next critical milestones?

Nikolaos: Flight 1 and its launch data review are behind us. We introduced processes to improve the systems that showed anomalies in Flight 1 and distilled all these learnings into concrete action points and new processes for the entire team. Flight 1 was a test flight.

Next is Flight 2 and shortly after that Flight 3, and it is during these flights that we aim to reach orbit for the first time. Demonstrate access to space and orbital insertion. After that, the goal is to achieve frequent and reliable launches. Very few companies reach orbit on their first or second try (SpaceX needed four with Falcon 1), so we are aware of the technical challenges that we could face in our second flight, as we enter into phases of the mission that we did not experience in Flight 1. “Fail-fast” doesn’t mean fail stupid, but sometimes another iteration is faster and perhaps even more cost-efficient than endless analysis.

Alex: I’d like to unpack the following quote from your company’s website: “Launching is the first step to space. By enabling access to space, we contribute to humanity’s progress and our planet’s sustainable technological and economic development”. What are the applications you’re personally most excited about by enabling space access?

Nikolaos: Our mottos are “launch for life” and “monitoring the world's most urgent issues”. Space is becoming a key platform for many industries worldwide, and many of them are utilising satellite-based technologies, often employing small to medium-sized satellites.

I’d say Earth observation and environmental monitoring are closest to my heart. Whether it's climate change tracking—such as CO2 monitoring or ice-melting and deforestation—or disaster monitoring for wildfires, floods, earthquakes, and hurricanes. Another vital area is agricultural optimisation, including moisture tracking and crop health.

Many of these applications already exist, but now we’re ramping up in terms of observation frequency and measurement quality. More frequent satellite passes. Higher resolution data. For example, when it comes to disaster monitoring and agricultural optimisation, these issues hit very close to home, especially when discussing Greece and the recurring wildfires we experience in the summer.

I'm pleased to see that, as a nation, we're developing our own satellite constellations for such applications. Earth observation using thermal infrared or synthetic aperture radar, addressing wildfires, floods, maritime safety, and environmental protection. I’m referring here to the National Microsatellite Program, which is part of the Greece 2.0 plan.

Seeing this develop, knowing it’s both a demanding application and one that Greece truly needs, fills me with great pride. We're trying to become the first nation with a sovereign satellite-based wildfire warning system. It will be of tremendous help, especially in the summer.

Alex: Even though we have a long history of aerospace engineering in Europe with the Spitfire, Concorde, Airbus, and more, why do you think we don’t have a reliable European launch provider yet?

Nikolaos: For years, Europe has relied on its main launch provider, Ariane 5, and now Ariane 6 and Vega. The main issues have been delays, cost overruns, and a general lack of flexibility. Although these products are solid from a technical perspective, I’d argue that, because it has been somewhat of a monopoly and very state-driven through European space programs, there was little incentive for improvement. No push to adopt new technologies, reduce costs, or increase performance. Compared to SpaceX, existing European rockets are more expensive and less agile, especially for commercial customers, who are price-sensitive and hence often do not pose a competitive alternative in the market. It is enough to look at the number of orbital launches for last year: Europe launched 3 times, whereas the US did it close to 150, and China almost 70 times.

Therefore, we lack a low-cost, high-cadence launcher, which is something we aim to provide. In short, the state-driven model didn’t reward innovation. Furthermore, the development of these vehicles has been influenced by geo-return policies (this refers to the European Space Agency's practice of ensuring that Member States receive contracts in proportion to their financial contributions to ESA programs). That benefits smaller states, but it creates fragmentation with more interfaces, slower processes, and inefficiency. That’s a key reason why Europe has remained stagnant in launch services for years.

Europe undoubtedly possesses the talent—both in software and hardware—and a robust industrial base. It’s not about a lack of skill. It’s been a lack of structural incentive to innovate due to the monopoly and the funding model.

Alex: I also want to zoom in on your own journey. How does one become a Chief Engineer at a rocket company?

Nikolaos: Space has always been a passion of mine. Since school, I’ve spent countless evenings immersed in astronomy and physics books, learning about space missions. I knew I wanted to work in space, either in space engineering or astrophysics. Eventually, all the dots connected.

Right after high school in Greece, I received a scholarship and moved to Germany for my studies. I began with aerospace engineering and later added physics, completing two degrees: aerospace (rockets and space missions) and physics (astrophysics and plasma physics).

I then pursued a master’s and PhD in aerospace engineering, with a focus on propulsion systems. My research employed both numerical and experimental methods to gain a deeper understanding of the extreme environments encountered in rocket engine combustion. That tied directly into my first role at Isar Aerospace.

I completed my PhD at the Technical University of Munich, where Isar Aerospace was founded (2018). At the time the company started, I was completing my final year of PhD studies at Stanford, so I didn’t join immediately. But as soon as I returned to Germany, I joined the team as a Propulsion Engineer, defining the architecture of the upper-stage engine.

After about a year, I started leading the combustion team. It was a challenging but incredibly rewarding experience, developing an engine from scratch, testing it in Arctic conditions at -35 to -40°C, going through tough iterations, and seeing it work. Knowing that your design is going to fly is an incredible feeling.

Last September, I transitioned into the role of Chief Engineer, defining the technical roadmap and overseeing the overall system architecture to ensure subsystems integrate cleanly and remove technical silos between departments. It's about ensuring the entire tech organisation operates as a cohesive system. You certainly need strong technical experience for this role. That’s non-negotiable. But you also need holistic, system-level thinking, balancing trade-offs between competing technical solutions, and understanding risk. And you need to align that risk with the company’s overall strategy.

Regardless of level (intern to CTO), what we value most is ownership and accountability. We’ve seen that prioritising ownership sometimes matters more than pure technical skill. If you own your system, there should be no one else in the company—or the world—who knows it better than you. From design to manufacturing to testing, you need to understand how your system interacts with upstream and downstream components. That kind of ownership is what makes a great engineer or technician.

Alex: Do you think a PhD is necessary for this line of work and space technology in general?

Nikolaos: As someone with a PhD, I’ve thought about this a lot. The short answer is no, a PhD isn’t necessary. I was very passionate about research, and early on I had to choose between academia and industry. I loved teaching and doing research, but I opted for the startup world. The skills you develop during a PhD, like deep problem-solving and specialisation, are very valuable. But they can also be narrow.

Some PhD holders transfer those skills well and make a big impact. Others are highly specialised and don’t immediately translate into something useful for a startup at its current stage. Meanwhile, we’ve had bachelor’s or master’s graduates with a few years of experience who brought immediate value.

So I’d say a PhD isn’t essential, especially in an early-stage company. It depends more on the person. If you do have a PhD, you need to be able to distil your expertise down to what matters and apply it in a practical, fast-paced environment. The key is finding the right balance between technical rigour and speed. Know what matters and what doesn’t. That’s what makes someone thrive in this industry.

Alex: That was so much fun and inspiring, Nikola! Thank you.

Share

Jobs

Here’s the list of startup jobs in Greece

News

Natech Banking Solutions raised $33m Series B ($26m in equity + $7m debt) ahead of the launch of Greek digital bank.

VODA.ai secured a Series A led by CRH Ventures, with participation from repeat investor L-Stone Capital, to help water utility companies manage and maintain their water pipes more efficiently using AI. To date, VODA.ai has analysed over one million miles of pipe across 26 U.S. states and six countries.

Istios Health raised $5.6m Seed led by BIP Ventures to deliver virtual infectious disease care.

RTDT raised $4m Pre Seed for its wind energy optimisation technology.

Voice agents for healthcare administration, Auxilis AI, secured funding from Genesis Ventures.

Maritime digital firm Vsltec secured €540k in Pre Seed backed by prominent Greek shipping personalities.

Technology solutions Qualco Group acquired majority stake in cybersecurity company Cenobe.

The Student Innovation and Entrepreneurship Competition, organised by the Archimedes Centre of NKUA, is accepting applications.

The Hellenic Centre for Defense Innovation (HCDI) has issued two new calls for projects.

Greece’s first nanosatellite DUTHsat-2 is now in orbit as part of the “GR Cubesats In Orbit Validation” program.

New coworking space SuiHub Athens by Sui. Some more coverage on why Mysten Labs (the company behind Sui blockchain) is doubling down in Greece: Link1, Link2.

Resources

All about the Greek startup industry w/ Panos Papadopoulos, Partner at Marathon Venture Capital.

The rise of autonomous science w/ Filippos Tourlomousis, founder & CEO at Biological Lattice Industries.

The role of nuclear energy in shipping by the Deon Policy Institute.

How to lead with product intuition, not just data w/ Renia Rigopoulou, Group Product Manager at Orfium.

Local governance meets AI w/ Christos Porios, founder of OpenCouncil and Schema Labs.

Building a fintech startup w/ Alex Christodoulakis, co-founder & CEO at Wealthyhood.

AI and robotics w/ Minas Liarokapis, co-founder & CEO at Acumino.

Reconditioning battery modules from discarded batteries w/ Aimilios Orfanos, founder & CEO at Watt4Ever.

Events

MIT Universal AI Summit by StartSmart South Eastern Europe on Jul 8

L&D Hub meetup #3 on Jul 15

ProductTank Athens meetup #9 on Jul 16

4th Athens Causal Data Science Meetup on Jul 17

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece—it helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X. Thank you for reading.

Alex

Subscribe now

Human or Bot?

Today’s newsletter guest is Anastasios Andronidis, Senior Staff Software Engineer at World, a new kind of digital infrastructure to verify humans in the age of AI. Since 2019, the project has raised close to $400 million. Hang on, as we'll delve into the specifics in a minute. With Anastasios, we also chatted about internships (he has interned seven(!) times in his career at places like Microsoft, Box, CERN, etc), how AI is changing the game for young engineers, and more.

Alex: We've got a packed agenda today. You're deeply involved in creating a technology described by co-founder Sam Altman as aiming to "uniquely identify and authenticate humans in the age of AGI." What exactly is World, and why do we urgently need a technology that verifies humans?

Anastasios: Imagine scrolling through Facebook or Reddit and finding a post with thousands of likes. How many of those reactions come from unique, real people, and how many from one malicious actor wielding a swarm of fake accounts to sway your opinion?

Over the past decade, artificial intelligence has leapt forward. Machines can now think, reason, and even outperform human experts. Each new AI benchmark is quickly surpassed. Systems that write and speak like people already exist, making it harder than ever to know whether what you read, watch, or hear online was created by a person or by AI.

Soon, advanced AI will gain agency: it will not only answer prompts but also act autonomously in the world. Although powerful and helpful, this capability can be abused. Picture your personal AI assistant calling a restaurant to reserve a table. Convenient. Now imagine someone unleashing dozens of assistants to book every table and hurt that restaurant’s business. To prevent such abuse, AI agents must prove they act for a single, real human.

Until now, we fought fraud with passport or national ID checks (KYC), a process that excludes many people and exposes private data. This is why in 2019, Sam Altman and Alex Blania launched World (formerly Worldcoin) to solve this in a more inclusive, privacy‑first way. World is building a human‑verification layer for the internet—World ID—that uniquely distinguishes real individuals from bots and AI agents.

Think of World ID like a Google or Facebook login, but with one crucial twist: you can have only one World ID. Clicking “Sign in with World ID” lets any online service confirm that each account belongs to one real person while protecting that person’s privacy. World doesn’t know where, when, or how you use your World ID, and the services you log into can’t tell that you’re the same user elsewhere.

World has already partnered with Razer (gaming) and Match Group (dating), allowing users to sign in with World ID to curb cheating, scams, and bots.

But how do you uniquely distinguish nine billion humans in an inclusive, private and spoof-resistant way? Traditional IDs aren't inclusive, as many people worldwide lack passports or credit cards, especially in rural or underserved regions. And these IDs can now be convincingly faked with AI. Face can be easily manipulated with makeup and beards. Fingerprints are not that inclusive, people who do intense manual labour lose their fingerprints over time.

World solves this by turning to your iris. The intricate, stable patterns in the iris are widely regarded as the gold standard in biometric security. By choosing iris scanning, World provides a simple yet highly robust way to verify each person once—and only once.

Privacy is foundational. World is building an open‑source, privacy‑first digital identity backed by zero‑knowledge proofs, anonymous multi‑party computation, and zero‑knowledge machine learning. Unlike mainstream login providers such as Google or Apple, World never tracks which online services its users access.

Alex: Can you walk me through exactly how verification works? If I’m not mistaken, the company claims to have verified 13 million people worldwide so far.

Anastasios: Happy to delve into the nitty-gritty technical details, as I have developed a significant portion of this. We use the Orb, an IoT device with the shape of a sphere about the size of a beach ball. The Orb houses multiple high‑definition, multispectral cameras and runs several neural networks locally to confirm you are a live human (though not a unique one) through liveness tests that resist makeup, photos, masks, contact lenses, and others. You simply look into the lens; within seconds, the device converts the captured images into an iris code, a numerical abstraction of your iris texture. We ensure the entire process is secure, completely local to the Orb and implements strong anti-spoofing and anti-tempering measures. I highly recommend that the curious readers look at our paper for more information.

The encrypted iris code is then transferred—end‑to‑end encrypted—to your phone, where you remain the sole data owner. The World App derives three unlinkable shards from the code and sends them to an Anonymous Multi‑Party Computation (AMPC) network run by three independent universities. None of these shards, on their own, can be linked back to the original iris code. The AMPC merely checks whether the combination already exists; if it does, registration is refused to enforce “one‑person‑one‑ID.”

In essence, the AMPC is a distributed encrypted database that checks if your Iris code has already been registered. The AMPC is split into three and operated by different universities. This system represents a significant innovation from our side, setting a new standard for privacy and security that no other company has achieved.

If the AMPC has no entry for your Iris code, then it allows you to register a unique World ID on the Ethereum blockchain. It is important to highlight that the AMPC doesn’t know which World ID you are registering on the blockchain, and this is how we guarantee that your Iris code information is completely independent of your World ID account. What this amounts to is a network of verified users tied to wallet codes rather than real names and a database filled with hashes rather than actual iris or other biometric photos.

When the registration is done, some Worldcoin tokens (World’s native cryptocurrency) are automatically transferred to your digital wallet as a reward for becoming a verified human.

If you were to use your World ID to access a website, that site would learn nothing about you except that you’re human (utilising zero-knowledge proofs). World also has no way to find where and when you have used your World ID, and lastly, different websites that you use your World ID cannot know that you are the same person across different services. The Orb is open-source and open-hardware, so outside experts can examine its code and verify our privacy claims. You can find more information about our audits here.

Alex: Why integrate a cryptocurrency into an identity network? How do identity and finance intersect here?

Anastasios: We believe a global network of verified humans shouldn't be controlled by a single entity, should be permissionless and self-sustainable. Cryptocurrency decentralises control, enabling community ownership. As a matter of fact, other companies can contribute to the development of the network, even build their own Orbs. Though still early, we are exploring how Worldcoin token holders will collectively govern the network, with each person holding equal voting power, regardless of the quantity of tokens held, ensuring genuine democracy through our unique verification process.

Alex: How do you feel about the criticism you have faced over the years? I believe the company has already been suspended from using the Orb in certain countries due to allegations of regulatory violations. Is it a peek into the future or a scene straight out of Black Mirror (British sci-fi TV series)?

Anastasios: World is a new primitive, something that has never been built before. While we open-source our technology so that everyone can examine the validity of our claims, understandably, some people (regulators included) don’t understand the technical details, and of course, on our side, we also need to communicate all this complexity better. Numerous clickbait articles online further fuel the confusion.

This happens with every new technology, right? Some level of criticism is always helpful, of course, and we take constructive feedback seriously. Yet, I’m proud that we've ignited critical conversations on privacy-first identity verification. These conversations are important for our future as KYC with sensitive information doesn’t scale, and it’s incredibly dangerous when data is leaked.

Alex: I want to shift gears before we wrap this up. You’ve had a fascinating journey from CERN to PhD to Big Tech to founding a startup before joining World as employee #50. Where would the 22-year-old Anastasios start from today?

Anastasios: I come from a small village near the city of Serres, in northern Greece. I attended music school in Serres and then went on to the Aristotle University of Thessaloniki. I’m very grateful for the opportunities the tech industry gave me as a software engineer in the 2010s and 2020s.

Nevertheless, I’m concerned that future computer science grads won't have the same chances. The advent of AI has already shrunk the pool of traditional available job openings for junior engineers and interns, and I expect things to get worse, given the leaps of progress AI is making. I receive the same feedback from engineering managers across big tech and startups: AI development tools have effectively replaced a significant portion of the work they used to assign to junior engineers.

Before I sound like a doomer, I want to highlight that AI is actually reshaping these traditional roles to something new. Writing code is not anymore the privilege of the elite. With AI, coding is a commodity. So now, to be a good engineer, you need to employ new attributes to yourself.

In that environment, I urge university students to work on side projects to stand out when the time comes to apply for their first jobs; Arduino projects, websites, etc. In the digital economy of the future, where we’ll rely even more on AI, employers increasingly prioritise creativity, entrepreneurial and doer mentality over academic credentials.

Alex: You’ve done a fair number of internships in your career :) Microsoft, Box, Facebook, CERN, etc. What’s the best way to structure an internship to make it useful for both the intern and the employer?

Anastasios: Yes, I was a professional intern early on — I’ve completed seven internships! AI hasn't fully overtaken human contribution yet, so internships still remain invaluable. There are several side quests we can assign interns to, which are not on the critical path of workstreams but are valuable enough to serve as a win-win for the intern and the employer.

I prefer 6-month internships, which give ample time for the intern to onboard, learn new skills, and deliver tangible results. At World, I've personally hired seven interns from Greece. I guide them by setting the vision and direction, rather than handing out rigid task lists, creating an environment where they learn to think like entrepreneurs and product-minded engineers instead of just technical coders. I’ll soon be opening another intern role, so frequently check our careers page: world.org/careers.

Alex: That was fantastic. Thank you so much.

Anastasis: Appreciate it, Alex.

Share

Jobs

Here’s the list of startup jobs in Greece

News

We got a fair number of Seed rounds last week:

• $5.5m for Typedef (generative AI workloads) led by Pear VC.

• €2.1m for Billys (property management) by Deep Capital Group, Genesis Ventures, Lambda Development, and others.

• €1.05m for Plan(e)t Foods (plant-based food) by Sporos Platform.

• €1m for Koios Care (AI & digital biomarkers for Parkinson’s) led by Evercurious VC.

• €1m for Bitloops (AI coding tool) led by Eleven Ventures.

• €1m for Mobito (mobility data) by Loggerhead Ventures, Sporos Platform, and others.

• €600k angel round for Tasteat (food delivery for travellers).

• €420k for commerce platforms MomCycle & StyleCycle by Sporos Platform.

Plus, a new fintech-focused solo GP fund Rerail with a $20m size, and an $8m Series A round for clean energy marketplace Station A. There was also a strategic investment in technology services Moro Technology by the Toyota Group.

Resources

Building a defence prime to protect Europe in the era of autonomous warfare w/ Dimitrios Kottas, founder & CEO of Delian Alliance Industries. btw, Delian is hiring for several roles across Athens & London, check them out here.

A fascinating fireside chat w/ Anastasis Germanidis, co-founder & CTO of Runway, on building the most successful AI media company of our days.

Elasticsearch and remote work w/ Panagiotis Moustafellos, Distinguished Engineer at Elastic.

Challenging the status quo in shipping w/ Antonis Malaxianakis, founder & CEO of Harbor Lab.

Taking a student project to Hollywood and the story of Krotos w/ Orfeas Boteas, founder & CEO of Krotos.

From 0 to 30 million users and the journey of Schoox w/ Lefteris Ntouanoglou, founder & CEO of Schoox.

Building Greece's DARPA by The Greek Analyst.

Dealing with tech due diligence by Christos Chatzis, Sr Technology Manager at Kaizen Gaming.

The beauty of vertical SaaS w/ John Melas-Kyriazi, founder & CEO of Standard Metrics.

Events

Building Less-Code and Real-Time Applications with Cognibase by Thessaloniki .NET Meetup on Jun 20

Mergers & Acquisitions Summit by Money Review on Jun 23

Automating EC2 Lifecycle Management with Terraform & AWS Systems Manager by AWS User Group Thessaloniki on Jun 24

Scale-up Greece meetup presenting Hack The Box on Jun 25

Elixir.Greece.thessaloniki/1 on Jun 26

Founders Arena by EnvolveXL on Jun 26

Automation and AI & Become a Quality Leader & Vitest Ministry of Testing Athens on Jun 26

Συνάντηση 2025.06 by Athens Python Users Meetup on Jun 26

MIT Universal AI Summit on Jul 8

Subscribe now

That’s all for today. Find me on LinkedIn or X.

Back in your inbox in two weeks with more Greek tech stuff.

Thank you for reading.

Alex

Subscribe now

Embracing Uncertainty

Uncertainty is everywhere in startups: people, technology, product, market. But this isn't just a nuisance founders can't avoid. It's essential to startup success. Initially lacking traditional competitive advantages like scale or brand, startups rely on uncertainty as a protection against incumbents and new entrants. Uncertainty becomes their initial moat.

It's 2017. Google dominates search: billions log on, type what they want, and advertisers pay top dollar to capture that intent. Holding ~90% of global search, Google rakes in more ad revenue than any company on earth.

In June that year, a landmark research paper in machine learning titled "Attention Is All You Need" is published by eight scientists working at Google. The paper introduces a new deep learning architecture known as the transformer, foreseeing the technique's potential for tasks such as question answering.

An AI startup founded two years earlier was quick to pounce. OpenAI relentlessly scaled this new architecture to build large language models like GPT-1 and its successors, leading them to one of the most successful commercial products in history. ChatGPT currently makes $415 million in revenue per month. Although Google saw this new technology first, it was a startup that capitalised on it to build the future of productivity.

Did Google simply just miss it?

While the technology deserves the praise it receives, its translation into a juggernaut business was far from certain until millions of people rushed to use its new consumer-friendly interface. In the higher echelons of Google, the project was seen as just another interesting AI project. Google found it advantageous to let others lead. As CEO Sundar Pichai explained: "It's not fully clear to me that it might have worked out as well. The fact is, we can do more after people had seen how it works".

There was… uncertainty.

Large hierarchical organisations prefer to wait for more proof to become available rather than face uncertainty. The tension between where decisions are made and where accountability is held pushes them to require data and rational explanations to approve actions: investment memos, financial models, discounted cash flow models, etc. How can you plan for something fundamentally uncertain? Most will ignore or quash it internally. This doesn't mean established companies never pursue highly uncertain projects, but they are far less likely to do so.

Many entrepreneurs find themselves in a similar position: dealing with significant uncertainties. Nevertheless, without a boss asking for rational explanations, they can go ahead anyway. The entrepreneur may decide to act, unlike the boss, because the entrepreneur doesn’t need to explain themselves to anyone.

The startup doesn't need to be the first to ever do what they're doing; OpenAI was not the first to experiment with transformers, nor was it the first to create a search platform. But either the category must be new enough that no company already dominates, or its product must be significantly better to grab market share from incumbents.

This doesn't mean everything about startups is uncertain. However, it's the nature of high-growth technology businesses that the most important value creation drivers must be uncertain. Until a moat is in place, only uncertainty will deter competitors, leaving enough odds for the handful that enter to succeed. Startups must choose to embrace uncertainty because most companies, particularly established ones, won't. Of course, these can't really be intractable problems because startups succeed in solving them every day.

Where cheap cash and startup playbooks have led to incrementalism and many talented people iterating on a tiny product surface area (and don't get me wrong, this can still yield success given how large tech has become), embracing uncertainty acts as a beacon of differentiation. We're always so biased trying to put things in a category, and not to think about what's unique, what's original, what's different. If it's already in YC's Requests for Startups, it's a little too late.

This brings me to another point... Differentiated startups, startups which use uncertainty as a moat, are the ideal vehicle for raising a community's tolerance for uncertainty. Everyone (entrepreneurs, employees, VCs, and the community at large) gets a crash course on ambition, risk-taking, and managing uncertainty.

I recently heard Vangelis Mihalopoulos, Yodeck's CEO, saying: "We had examples in our network, like Bugsense and Epignosis, that built global B2B SaaS products from Greece. If they could do it, why not us?". We connect most with people who walked the same school halls, grew up down the same streets, or speak the same language. The next generation of Greek founders will look up to Augmenta, Hack The Box, Delian, and others building where most see uncertainty.

It's 2016. Long before ESG turns into a buzzword, two engineers/farmers straight out of university in the small city of Volos, Dimitris Evangelopoulos and George Varvarelis, start retrofitting hardware and software in tractors to help fellow farmers increase their yield and minimise fertilisers. Their technology becomes used across thousands of acres globally and, via the acquisition of Augmenta <> CNH, is now an integral part of new tractors from the world's second-largest agricultural equipment company.

It's 2017. While hackers still learn how to hack through textbooks and assessed via multiple-choice Q&A, Haris Pylarinos starts a side project (initially named in Greek "χάκαρε το κουτί"), an online learning platform for cybersecurity enthusiasts with social features built in. Nine years later, Hack The Box is used by 2.7 million hackers and the largest organisations globally.

It's 2020. Dimitrios Kottas, an engineer at Apple in California, returns to Greece to build Europe's modern defense prime. This is long before ReArm Europe, the war in Ukraine, and the recent defense tech hype. Today, they have created a globally competitive autonomous anti-access/area denial solution that's orders of magnitude more cost-effective and scalable than legacy systems. See their recent video below:

Embracing uncertainty is primarily a cultural phenomenon, and Greece, shaped by its turbulent past, has become allergic to ambiguity.

But it hasn’t always been like that. I wrote about counterexamples in the "We need to 10x the entrepreneurs", like the shipping DNA on the island of Chios, where life (and risk) at sea is honoured and celebrated.

The collective change will start with those who ignore the mimetic warfare of the Gartner quadrant, the VC theses and startup requests, and think about what's unique, what's original, what's different. With uncertainty comes volatility, simultaneously enabling outsized success and idiosyncratic risk.

We only need a handful of shots on goal for the vibe shift to begin.

Share

Jobs

Here’s the list of startup jobs in Greece

News

This is a jaw-dropping Seed round for benchmarking for LLMs, LMArena. They raised $100m Seed by a16z, Lightspeed, Kleiner Perkins, and more.

Plenty are going after the “rebuild the CFO suite” vision, yet Rillet seems to be doing something right. They claim 5x YoY revenue growth for their AI-native ERP and just raised $25m in Series A by Sequoia.

Color-changing nanomaterials Quantum Light secured $1m Pre Seed to build the future of pigments and dyes.

Athens-based maritime workflow solutions CompassAir was acquired by Insight Partners-backed Sedna.

Next-gen surgical patch Veltist raised €160k to eliminate a major surgical complication responsible for thousands of deaths each year.

8 weeks incubation program Astylab Founders is accepting applications.

Athens is a Rising Star tech hub in Europe according to Dealroom.

Kos Biotechnology Partners, a new biotech fund is launching soon.

Satellites made in Greece by ICEYE.

Resources

Recording from the latest Open Coffee in Athens with the founders of LearnWorlds, ORamaVR, and Biological Lattice Industries.

Building the world leader in online corporate learning with Thanos Papangelis and Dimitris Tsingos, founders of Epignosis.

2025 HR and salary trends by Randstad.

Building a car rentals marketplace with Nikos Drandakis, founder of Sync.

Nuclear energy and the case of Greece with Achilleas Chekimoglou, author of the book “Atomic Era: Nuclear Energy, Reactors, and Uranium in Greece in the 20th Century”.

Promotion rejections, performance improvement plans, and other dreaded conversations by Christos Chatzis, Sr Technology Manager at Kaizen Gaming.

Events

Join us for a special Open Coffee Athens #124 on Jun 19 with Anastasis Germanidis, co-founder of Runway.

Sui under the hood part 2 by Sui Community on Jun 6.

Greeking Out in Paris by Endeavor on Jun 10.

Patras Tech Talk 2025.06 [Summer edition] on Jun 11.

Serverless Edition by AWS User Group Athens on Jun 12.

Structured Concurrency, concepts and applications by foobar io on Jun 17.

AI Meetup Athens by Starttech Ventures on Jun 18.

Tech Tour Investors Summit 2025 on Jun 19 - 20.

1st International Humanoid Olympiad by Acumino & Endeavor on Aug 29 - Sep 2.

Subscribe now

That’s all for this week. A lineup of special guests is coming up in the newsletter:

• Human or bot? w/ Anastasios Andronidis (World)

• Accelerating AI w/ Polychronis Xekalakis (Nvidia)

See you in two weeks.

Find me on LinkedIn or X. Thank you for reading.

Alex

• Are you a pirate? + introducing Marathon Fund III

• Startup. Exit. Repeat.

• The road to superintelligence

If you’re not a subscriber, hit the button below and come aboard:

Subscribe now

Scaling Consumer Marketplaces

Thalia Geladaki is the marketing mastermind behind two of the most successful consumer marketplaces (and brands in general) that started from this corner of the world: Beat (ride-hailing app acquired by Daimler in 2017) and Skroutz (e-commerce marketplace backed by CVC, one of the largest private equity funds). We chatted about:

• How she got into startups

• How Beat turned from a Greek brand into a Latin American powerhouse

• Lessons learned scaling a marketing team from 4 to 44

• The transition of Skroutz from a price comparison site to a full-blown e-commerce marketplace

• Expansion into new countries: strategies and tactics

• Advice for aspiring tech CMOs

Let’s get to it.

Alex: Thalia, you have led marketing for two of the most successful consumer marketplaces started in Greece. I want to start with a brief introduction and how you got to the CMO position at Skroutz.

Thalia: It was not a straight shot. As a student, I dreamed of an academic career in the History of the Arts. I left Greece to study in London and worked in museums and galleries. Then, returning to Athens, I jumped into digital & communications and spent a number of years in agencies like CIVITAS and FMCG (Coca-Cola) before entering the startup world.

I joined Taxibeat as the Head of Marketing in late 2016, when we were operational only in Greece and Peru. Months later, Daimler acquired us, and we rebranded to Beat, expanding across Latin America as a ride-hailing app.

This was a significant shift from how Taxibeat operated as a platform connecting passengers with taxi drivers in the early days. We were a tiny marketing team of two that rebranded everything — apps, SWAG, even the office. It was intense but thrilling. I became CMO as we scaled into Colombia and Chile, growing the marketing team from 4 to 44.

Alex: Let’s zoom in. You took Beat from a Greek brand to a Latin American powerhouse in a year. What was that like as CMO?

Thalia: It was a wild ride. In 2016, we were in Greece and Peru; by 2017, we added Chile and Colombia, with Mexico next.

The biggest hurdle? Cultural gaps. Latin America’s vibe is worlds apart from Europe. We adapted our messaging, campaigns, and even work styles to build rapport with local teams. We had the leadership team fly over from Athens to spend time there, hire fast, fire fast, and set the right foundations to scale.

Athens was our HQ, while Lima was our LATAM hub, with offices per country. Central teams set the strategy, from marketing to design to operations, and local teams tailored it, ensuring the right tone and channels, sometimes with the support of a local agency. Was the wording proper? What about the tone of voice? Was it the right growth channel?

It wasn’t an overnight success, but in the end, locals perceived us as a Latin American brand. This gave us a massive advantage over Uber, which had already become an industry giant without much localisation.

A standout moment was Peru’s FIFA World Cup 2018 qualification. We launched an emotional campaign, a short film with drivers and passengers sharing their pride, capped with “11 players represent 11 million” (Lima’s population). We created relevant merchandise, ads, etc. No hard sell, just a subtle Beat signature. It went viral. This was a great case of cultural insight that helped a foreign brand connect emotionally with the local population, cementing us as a market leader.

Alex: As Beat’s CMO, marketing grew from 4 to 44 people. What were your biggest lessons learned scaling that team?

Thalia: If I could summarise in a few words the main points, these would be:

a. Respect the cultural differences of teams in other countries.

b. Have frequent physical presence across offices.

c. Introduce processes to ensure everyone knows what’s expected of them and is aligned with the company goals.

Alex: Jumping from one success story to another… You left Beat and then joined Skroutz, Greece’s largest e-commerce marketplace, as the company’s Chief Marketing Officer. During your tenure, Skroutz transitioned from a price comparison site to a full-blown marketplace with delivery, payments, loyalty, etc. What drove that repositioning?

Thalia: The company had already planned to transition to a marketplace before the pandemic, but the whole idea was planned across 5-6 years. Then covid lockdowns happened, e-commerce boomed, and we had to deliver this vision much sooner.

Most people don’t know this, but Skroutz launched a marketplace in 2016 — the first marketplace product category was pharmacies — but the core business remained price comparison. The company didn’t need marketing, operations, merchandising, commercial, or partner support for a very long time. It was mainly product & engineering, and a small customer care team.

Then, as commercial activity surged during covid, the company built more functions to support the entire customer journey from search to purchase to delivery and after-sales. Most merchants also embraced this transition. About 3,000 stores without an online presence got onboarded for free during the pandemic. We also acquired companies like SendX (delivery) and Everypay (payments) to accelerate things. This is when my journey began at Skroutz in early 2021.

As a newly formed marketing team, our job was (and still is) to amplify the great work done by our engineering & product teams across all functions. Skroutz is first and foremost a technology company. For example, in 2021, we removed the hat from our logo; a big internal debate, as that was part of the brand’s identity for years, which, however, tied Skroutz to “cheap deals”.

We started communicating clearly inside our platform and across all digital channels that consumers could now buy products directly from us. We even launched our first TV advertising. Skroutz vans and pickup points made us visible on streets and in neighborhoods. I would say from 2021 to 2023, marketing’s primary role was to reintroduce Skroutz to our users.

Alex: What about expansion to other countries? Do you see similarities in the expansion strategy between Skroutz and Beat?

Thalia: In 2013, Skroutz launched in Turkey under the Alve brand. We believed that because Skroutz in Greece had a strong product and content, this would be enough to succeed in Turkey. It was very different culturally, with a small launch team based in Turkey and no marketing or substantial investment. Eventually, the company had to pull out of this market 7 years later. The UK in 2015 was another expansion attempt that lasted almost 4 years.

But then, Cyprus was the first market to launch as a marketplace three years ago. We share the same language, culture, plus location proximity helped a lot. In the first years, all marketing activities started from our central team in Greece with the help of a local agency before hiring a Marketing & Operations manager in Cyprus. In the coming months, we will also be launching a Cypriot branch of our delivery company, Skroutz Last Mile (SLM), on the island, and we plan to install over 50 Skroutz Lockers in key locations across Cyprus, enabling faster and more convenient access to products for communities throughout the island.

We also launched in Romania ten months ago due to its proximity to Greece and because it’s a very large market where e-commerce is booming. There, our approach has been quite different. We silently launched and are taking enough time to understand whether our offering works while only using email and performance marketing like Google ads. It’s the opposite of Beat’s LATAM push, but this also reflects the product and culture differences between the two companies. Beat was more aggressive in hitting the goals.

We are a team that loves building, and experimentation is part of our DNA, while at the same time, we are humble and can have bias for action when needed, but also take the time to do things right. This year, Skroutz turned 20 years old, and since January, we have been running our biggest marketing campaign so far, using the reviews that users have left us all these years. A campaign inspired by the good and not-so-good user reviews and crafted entirely by our in-house marketing team.

Alex: Before we wrap things up, what advice would you give aspiring tech CMOs?

Thalia: I would say two things:

a. Marketing and communications should adapt to the business needs and not vice versa. Their only north star and goal should be helping the business grow, and in the case of technology companies, it’s about surfacing the product & engineering work.

b. And the second is linked to how I have approached my career. Looking backwards, most of my decisions could be attributed to me trying to maximise the learning curve in every new role and loving the company cultures and brands. I think the latter is vital for every marketer.

Alex: Thalia, thanks for sharing your journey here!

Thalia: My pleasure, Alex.

Share

Jobs

A fresh list of 400 startup jobs in Greece! Check it out here.

News

Axelar (interoperability for blockchain) secured $30m from several crypto investment funds.

Protio (proptech) raised €1m from Loggerhead Ventures, Lars Rasmussen, Genesis Ventures, and others.

Wocap (working capital marketplace) announced €450k pre-seed led by Genesis Ventures.

ICEYE (space tech) opens a new office in Athens to create a satellite assembly production line.

JPMorgan to withdraw UK lawsuit against Viva Wallet (neobank) but will continue pursuing legal action in Greece.

Watt4Ever (repurposing discarded EV batteries) was acquired by Sortbat.

ParkZen (real-time parking availability) was acquired by Parking Guidance Systems.

Resources

Video recordings from StrictlyVC Athens: a roundtable conversation on “Why Greece, why now: Betting big on the country's next round of founders” and “Greece's big tech moment” with Greece’s Prime Minister.

Building a digital health startup with Joseph Alvertis and Konstantinos Gkovedaros, founders of Nudge Care.

What open source adoption looks like in 2025 with Manos Koukoumidis, founder & CEO of Oumi.

Reimagining customer support with AI with Alex Marantelos, co-founder & CEO of Intryc.

On fundraising by Ilias Louis Hatzis, founder of Inkjin.

From triremes to drones: Why innovation alone won’t win a war by Stavros Messinis, founder of Smart Flying Machines.

Events

Open Coffee Athens #123 on May 30. Join us!

Reliable Testing Beyond E2E by Thessaloniki Software Testing and QA on May 27

Συνάντηση 2025.05 by Athens Python Users Meetup on May 28

Innovation Northern Greece 6.0: AI the day after on May 28-29

7th Athens eCommerce MeetUp on May 29

UX practices at Skroutz by Athens UX Community on Jun 4

LeadDev Athens Meetup #1 on Jun 5

Silicon Valley Greek BioHub | Meet-up #2 on Jun 6

Subscribe now

That’s all for this week! Find me on LinkedIn or X.

Thank you for reading.

Alex

Subscribe now

Are You A Pirate?

First things first, Marathon is back for Fund III: a €75 million fund.

We double down on what we set out to achieve since 2017. Same mission, stronger ambition.

• Invest in founders who address impactful, necessary innovations with a shared responsibility to shape a better future.

• Bring the global Greeks in tech community together and help build world-class companies.

• A close-knit venture team that rolls up its sleeves to support the next generation of founders.

Here’s some personal news, too. After joining Marathon almost five years ago, I’m excited to be stepping up as a Partner. I will dig a little deeper today, pulling back the curtain on our philosophy, focus, and my own journey.

Let’s dive in.

Purpose-Driven, Founder-Centric

We back exceptional founders building the future. This motto encompasses most of our investment decisions, which helped us capture the current zeitgeist well before AI, climate, robotics or defense became the norm.

We are not focusing on any specific sector per se. The best founders will always be three levels deeper and three steps ahead. Which means, it’s all about identifying exceptional people solving important problems and winning the right to work with them. Investment theses (just like startup advice, as Matt Turck recently wrote on X) are often terribly wrong and bound for a strong reality check.

Don’t do services → Palantir

Don’t build hardware → SpaceX, Nvidia, Tesla

Don’t sell to the government → Anduril

Instead, we are focusing on people changing their sectors versus chasing trends. These are only a few examples:

• Augmenta: A hard-core climate tech startup that offered a new sustainable way to farm lands. The company was acquired by CNH, a global leader in agriculture technology, for $110 million.

• Causaly: The most advanced AI platform for scientific discovery, providing research teams with comprehensive biomedical knowledge for drug discovery. Raised ~$100 million and trusted by some of the largest pharma companies.

• Hack The Box: The #1 cybersecurity upskilling platform loved by millions of hackers and thousands of organisations.

• SMPnet: Power grid software helping grid operators manage their networks and address the challenges caused by the intermittency of renewables (the core reason behind the massive electricity outage last week in Spain).

• Delian Alliance Industries: A modern defense prime to protect Europe and its allies using autonomy at its core. Started by an ex-Apple and Palantir founding team long before the Ukraine war.

• PolyModels Hub: The digital backbone to model critical drug development processes. In less than 12 months of ops, they are helping several top10 pharma optimise their blockbuster drugs.

Doubling down on our community, plus more

There are more opportunities than ever to build or join the craziest, most ambitious startups. Greece now has its first unicorns, hundreds of millions of dollars invested yearly, and 100+ exits in the past years. Many companies leverage local talent to serve leading global customers and markets from day one. Along the way, the community has been accumulating experience and expertise, while the successful generation of founders is taking an active role in giving back to the next generation of builders.

In the past, I had newsletter guests building from nuclear fusion reactors to superintelligent agents, and AI-designed drugs to blockchains with billions of dollars in transactions. Many such interesting conversations originate from the Greek diaspora; London, Zurich, Munich, New York, etc. As a matter of fact, we have put together Greeks in Tech events in dozens of cities from San Francisco to Tokyo and half of the companies we have invested in started outside Greece.

Our community has been taking us places, and with our third fund, we will also be looking for opportunities that don’t necessarily come with a Greek tag. Over the years, we have experienced a lack of investors with the courage to invest in unpopular markets early on and move fast with conviction across Europe. These might sound like table stakes; they still are not.

Europe lacks founders or operators turned investors. In Silicon Valley, more than half the investors at top venture firms have worked at or run a startup themselves. In Europe, this figure is a pathetic 8%. We believe founders and operators looking for meaningful missions to put their experience and energy behind are what the next generation of European founders deserves.

Stepping up as Partner

George and Panos made a big bet on me in the middle of 2020. I had just started an anonymous Twitter persona, writing about Greek startups and how I viewed the tech world as a failed founder turned product manager. It was a fun little side project.

I worked for startups and built stuff before. It all started in a London “hacker house” in the early 2010s, where we got from a quiz mobile game — QuizUp was all the rage then — to an ML ops platform that secured a winning spot in one of the largest hackathons (4,500 teams organised by ProductHunt), before joining two prime examples of companies starting from Greece to become global category leaders (Workable and Blueground).

But I knew zero about startup investing. Yet, George and Panos offered me a seat in their close-knit group of community-first venture builders and got to see from up-close the journey of a number of world-class companies emerging from that same community I had been part of for the past decade.

It's been incredible to grow alongside these stellar people, and I'm grateful for having the chance to step up as a Partner to continue building Marathon together and work with the next generation of ambitious founders building the future. We will each take on a tiny number of projects each year – 1-2 typically – and we will do everything in our power to further the founders’ mission.

New veins for Jack Sparrows

Startup Pirate originated from my realisation that the tech industry had to get its soul back as a pirate haven, not a navy port. VCs were becoming like bankers, optimising for assets under management and evaluating startups based on “who else is investing?”. Series A startups with avocado breakfasts and kombucha on tap in shiny WeWorks were becoming the norm. Not much has changed since (did you hear about this $2B Seed at $10B valuation pre-product, pre-revenue?)

Why did some people back in the 17th century, or whenever, become pirates? The likely payoff was abysmal, I imagine. A slight chance you’d make a fortune from some prize and a large chance you’d drown or get shot. Unlike most other people, they lusted after that risk. The potential for riches was an argument for the venture. But the real payoff was the pirate life itself.

And today, more than ever, the tech industry needs Jack Sparrows, running around uncharted waters, with small trusted crews, great sailors evading the navies, an unreasonable desire for gold and for having a great time. Entrepreneurs who work hard with their heads down, are capital efficient, take risks, and venture into the unknown to build their craziest, most ambitious ideas.

And here I am — with a small bag of gold looking to help the real pirates do real pirate stuff! Drop me a line to chat ;)

🏴‍☠️

Share

Jobs

A list of startup jobs in Greece! Check it out here.

News

Unblocked (contextual code intelligence) raised $20m Series A

Astrato Analytics (business intelligence platform) raised $5m led by Big Pi Ventures

Acromove (private 5G & edge cloud data center) raised €3m by EFA GROUP

Dex (AI recruitment) announced a $3.1m Pre Seed led by a16z

Coffeeco Upcycle (reducing coffee waste) secured €715k

Qualco (fintech in credit management) will IPO on the Athens Stock Exchange

Another large NASDAQ semiconductor company, Qualcomm, has an R&D presence in Greece

Resources

The PM role is dead. Long live the PM. by Joseph Alvertis, founder of Nudge Care

A customer support retrospective by Anastasia Tatarinoff, Director of Customer & Partner Support at Skroutz

Open source software and entrepreneurship with Costas Tsaousis, founder & CEO of Netdata

Notes on Cursor and AI development by Spiros Martzoukos, Senior Product Engineer at Findbar

The role of Analytics Engineer with Panagiotis Apostolopoulos, Analytics Engineer at Hack The Box

Creating an AI-powered knowledge base for your team by Pavlos Petros Tournaris, Sr. Staff Software Engineer at Blueground

Events

Day 3 of Panathenea festival on May 9

28o Open Coffee Heraklion on May 10

Greeking out in London by Endeavor Greece & Hellenic Tech Network on May 14

Greeking Out in Cambridge by CJBS, Big Pi Ventures & Endeavor Greece on May 15

L&D Hub meetup #2 on May 15

Patras Tech Talk 2025.05 on May 16

Elixir.Greece.Athens.meetup/1 on May 16

GCC meets SUI by Greek Cryptocurrency Community on May 17

Innovations in Medical Technology by Hellenic Tech Network on May 20

Emerging Managers Summit by Startup Pathways on May 21

Athens WordPress Meetup on May 21

Agentic AI for developers and vibe coders by PyData Athens on May 22

Take Back the City #7 by Astylab on May 27

Subscribe now

That’s all for this week, and see you in two weeks!

Find me on LinkedIn or X.

Thanks for reading,

Alex

As I previously wrote in this piece, rebellion, creativity, and risk-taking are culturally ingrained. It’s watching your peers venture and succeed or fail. Startups cannot be taught nor cultivated through tax credits & non-dilutive grants. If this newsletter helps at least a handful of people connect the dots and go build or join the craziest, most ambitious startups, I’m happy :)

Let's roll on to the decade milestone. If you’re not a subscriber, hit the button below and come aboard 🏴‍☠️

Subscribe now

Startup. Exit. Repeat.

The following is a conversation with Conno Christou, co-founder & CEO of Keragon, a workflow automation platform for the healthcare industry. Conno previously co-founded Avocarrot, an ad tech company acquired by Glispa Global Group in 2016 for $20 million. Conno shared his fascinating journey building startups since 2012:

• Why ad tech was broken in the early 2010s and the start of Avocarrot

• Getting acquired by a prominent ad tech company

• Persuading the acquirer to double down on the Athens office

• From ads to healthcare, and founding Keragon

• Fundraising and hiring as a repeat founder

• Missing the raw energy and naive optimism of a first-time founder

• AI as the game changer for team productivity

Let’s get to it.

Alex: Conno, it's fantastic to see you and the rest of the Avocarrot founding team back in building mode. Correct me if I'm wrong… You have known each other for 20 years, and your story hacking together started with a fashion application in 2012?

Conno: Oh, my days - that’s ancient history now! With George Eracleous, George Makkoulis, and Panos Papageorgiou, we all served in the military together in Cyprus, then attended the same university in London, and after graduating in 2012, we started going to hackathons and building stuff.

At some point, we developed a fashion application that allowed users to share photos of clothes they wanted to buy, and their friends could vote with a thumbs up or down. With that prototype, we won the London Hackathon. We explored ways to make money from this and realised mobile advertising was clunky, intrusive, and not optimised for user experience. We decided to build an advertising platform prioritising the UX on the app developer side - a novel idea for the early 2010s. Facebook and Instagram started serving native mobile ads 1-2 years after us. The aha moment was when we realised that click-through rates were 15x-20x higher than incumbent solutions; hence, we went all in on developing that technology. That's how Avocarrot was born. We got some small funding, which gave us runway for a few months while building the beta version across London and San Francisco.

Alex: Fast forward to 2016, when advertising tech company Glispa acquired you. How did that acquisition come about?

Conno: Throughout those years, we had built an ad exchange with an emphasis on the UX and the look and feel of the ad, while developing a real-time bidding engine in the background. This allowed advertisers to bid on that inventory and get in front of the right people at the right time, as we had gathered tons of user metadata. Around that time, we were live on 10,000 apps worldwide, serving approximately 4 billion ads per month. Our network and technology were valuable for larger players such as Glispa, which eventually acquired us in 2016. They were very bullish on our team, our tech, and the timing.

Alex: At that time, most of your engineering team was based in Athens. We have observed several instances where local teams remain flat or gradually fade after the acquisition. But that was not the case with you. Does it always come down to winning the internal politics game, and what were the main arguments for doubling down in Greece?

Conno: Moving to Greece to build the company was one of our best decisions. At some point in late 2013, we had to move back to Cyprus because the company was running out of money. Both London and San Francisco were very expensive cities to live in, and we still hadn’t raised that $2 million Seed round. After a few months of interviews, we decided to review more candidates for our team, so we ran parallel interviews in both Cyprus and Greece. We had our first exceptional hiring interview in Athens, and hence we decided to relocate there. This eventually worked out wonders. When Glispa acquired us, we were a team of 17, most of us based in Athens.

Company politics is a big part of the game. Our main negotiating power was one thing: always over-deliver. Since the early days of the acquisition, we have been continuously shipping features, driving revenue, and building trust. After Avocarrot, Glispa acquired another six or seven companies, while the leadership team continued to bring these products under our umbrella. Imagine the leverage we had by leading all of them. That was the primary driver for convincing Glispa to invest further in local engineering talent. Three years later, we were a team of 50 strong engineers in Athens.

Alex: How long did you stay around after the acquisition?

Conno: 3 years, 8 months, and 17 days.

Alex: You were really counting down the days, weren’t you? What drew you to healthcare and Keragon?

Conno: It took us a while to jump into another idea. All four of us left Glispa at the same time and took some time off to recharge and think of our next steps. That break lasted for about two years. Healthcare drew our attention early in our brainstorming for the following reasons: a. It’s a space we can positively influence people's lives, even indirectly, b. We believe healthcare is ripe for disruption in the next 5-10 years, and we wanted to ride that wave.

With Keragon, we are building the "Zapier for healthcare". We integrate with the most widely used healthcare vendors, enabling our clients - the doctors - to set up automations across different tools without needing code. Right after COVID, the healthcare industry was pushed to go digital, and a lot of VC money chased startups creating best-of-breed solutions for each part of the operation: scheduling, CRMs, patient intake forms, engagement platforms, etc. However, as of today, these platforms don’t connect with each other. Doctors typically use a dozen siloed new digital tools on average. Moreover, there's a shortage of 30,000 doctors in the US alone. Imagine how overworked they are. There's a lot of burnout and turnover. It's a mess. That's why we came in.

Currently, this problem is very acute, and the opportunity is huge. We solve this by providing a very intuitive platform with a no-code editor, for anyone to drag and drop the tools they use and set up workflows in an "if this, then that" manner. You set it up once, and it's automated from then on. The nuance here is that compliance, security, and reliability are of the utmost importance in healthcare. You are not automating just a Slack notification - these are patient workflows, in most cases clinical/critical operations for healthcare organisations. You need to be reliable, secure and compliant at scale. That’s why we chose to keep a low profile for two years before our public launch, to build internal confidence in these three critical pillars. The threshold for an MVP with Keragon was much higher than placing ads in mobile apps.

Alex: I want to switch gears and take you to company building, and specifically, go-to-market, as a second-time founder. I often think about the tweet below by Justin Kan, the founder of Twitch. Do you agree, and is it how you approach things now with Keragon?

Conno: Ha, first time I hear this - there’s definitely some truth to it. In our case, even though we’re still kind of obsessed with product, we now actively think about how distribution can be embedded directly into the product experience. That means being intentional about how new users find us, how word spreads, and how we reduce friction to adoption.

It’s a bit unusual in healthcare, where services often take the lead. But all four of us come from the tech world, so for better or worse, we naturally approach things through a product-first lens - even when we’re thinking about growth.

Alex: I'm curious to hear what you've done differently with fundraising and hiring this time with Keragon.

Conno: We are more deliberate now. Sure, being second-time founders with an exit under our belts made things easier for us. We raised our first round when we had enough conviction that this idea was worth pursuing. We were new to the space, so we initially took enough time to talk to potential customers, explore the healthcare technology ecosystem, and develop an early version of Keragon. We didn't go out to raise funding on day 1.

On top of that, we only contacted investors we thought would be a great match based on their experience and thesis. Since we’re new to the healthcare space, we sought to partner with someone who has a deep understanding of healthcare to complement our product and technical backgrounds. We have raised two rounds: $3 million and then $7.5 million, following initial customer traction. As everybody says, good traction makes everybody want to work with you. In later rounds (Series A+), the second-time founder effect diminishes as you’re now raising on a spreadsheet, not on a deck.

Hiring was one of the things we learned a ton about while building Avocarrot. We have always been ruthless about candidate fit, using three pillars to determine this: people who are smart, get things done, and have the right cultural fit. A candidate must tick all three boxes to be hired. We are fortunate to have worked with some fantastic individuals in our previous company, which provided us with a significant boost in the early days of Keragon as we joined forces with a few of them again.

Alex: There’s data to back up that second-time founders are often far more equipped to succeed than their inexperienced earlier selves. However, I want to turn the argument around. What are the “benefits” of first-time founders?

Conno: First-timers often possess raw energy and naive optimism, which are great traits for founders. There's magic in not knowing what's impossible! They jump in with fresh eyes. They may have heard about how challenging it is to build a company, but they don't truly know. First-time founders lack painful memories that introduce hesitation. And it's the lack of fear of failure that empowers them to pursue bold new ideas.

In contrast, second-time founders have experienced the ups and downs of their first startup and are aware of the risks and challenges. They have scars that can introduce hesitation, so they make excuses not to pursue something new, which is partly why it took us so long to decide on our next startup idea. Analysis turns into paralysis… On our end, we wanted to "hack" that naivety, therefore, we decided to venture into a new sector: healthcare. While brainstorming for new startup ideas, we had a very important rule: anything but ad tech.

Alex: We talked offline about this before… You are a strong proponent of using AI internally at Keragon to boost your team's productivity. I'd love to share your thesis about that. For context, I'm also attaching the recent memo from Shopify, pulling some quotes from it: "Using AI effectively is now a fundamental expectation of everyone at Shopify", or "Before asking for more headcount and resources, teams must demonstrate why they cannot get what they want done using AI".

Conno: We continuously try to find ways to 10x the productivity of our team. We are reshaping the company through processes and projects to force us to adopt AI tools and become an AI-first company. This Shopify memo is spot on. We know we are leaving a lot of optimisation on the table, and it's something we are doubling down on. It's actually one of our company's five core values: If you find yourself doing something manual twice, automate it. We always think about automation with AI at the core. We believe we can be 10x more productive using automations and AI. We’re an automations company after all..

Alex: You mentioned five core company values, so I have to ask… What are the other four, apart from automate things?

Conno: a. Over-delivering (the team expects each of us to work both hard and smart); b. Over-communication (crucial for a hybrid team like us); c. do it yourself (signals how we value ownership of work and taking initiatives); and last but not least, d. don't be an asshole :)

Alex: Conno, thanks a ton!

Conno: Alex, really enjoyed this chat.

Share

Jobs

A fresh list of startup jobs in Greece! Check it out here.

News

Endor Labs (software supply chain security) raised $93m Series B led by DFJ Growth.

Plum (fintech) secured £15m in venture debt from BBVA.

Prosperty (proptech) raised €5.5m Series A led by Eurobank.

Mood (music event discovery) announced an angel round of €1m.

Bolsterup (construction tech) raised €500k led by UniFund.

European Dynamics (IT service provider) raised funding.

Decode+ (legal AI) secured funding from Genesis Ventures.

Couch Heroes (gaming) announced a €110k round by Alter Ego Ventures.

Kos Biotechnology Partners is a new investment firm in Greece focusing on biotech and life sciences.

Greece’s national stock exchange assesses the upgrade of its electronic book building to blockchain using Sui.

Resources

Coding with AI: The end of software development as we know it with Panos Papadopoulos, Partner at Marathon, and Gerasimos Tsiamalos, co-founder of CSSIgniter.

Lessons learned from $1 billion in course sales and 23 million learners by Panos Siozos, co-founder & CEO of LearnWorlds.

To pivot or to not? by Dimitris Nikolaou, co-founder of Wondercraft.

What DoorDash got right and most others missed by Babis Makrynikolas, SVP Product & Pricing at Blueground.

An introduction to robotic surgical devices with Manios Dimitrakakis, founder & CEO of Panda Surgical.

Nuclear energy and SMRs by Georgios Laskaris, Nuclear physicist and President of Deon Policy Institute.

UX & Marketing: Balancing business goals with Michail Kouroupakis, Director of User Experience at efood.

The role of the Data Architect with Christina Kalimeri, BI & Data Analytics Manager at Cardlink.

Why centralising marketing in the AI era could backfire by Giorgos Vareloglou, founder & Managing Partner at REBORRN.

Startups don’t die. Founders quit. by Ilias Louis Hatzis, founder of Inkjin.

Events

Open Coffee Thessaloniki #87 on Apr 25

Entrepreneurship and Growth: The future perspective by NBG on Apr 28

Ctrl+Alt+Migrate: Rebooting to the cloud by AWS User Group Athens on Apr 29

Angular Athens 26th Meetup on May 6

StrictlyVC Greece on May 8

From legacy Java application to microservices by Java Athens Meetup on May 8

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece—it helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X.

Thanks for reading,

Alex

• From 0 to 60,000 customers

• Are we getting AI-powered drugs soon?

• GPS and electronic warfare

If you’re not a subscriber, hit the button below and come aboard:

Subscribe now

The Road to Superintelligence

“When you are young, take as many risks in your career as possible. Join the craziest, most ambitious startups because this is where most breakthroughs happen. This is the best way to work on things that matter and have an impact.”

Ioannis Antonoglou is one of the very few people who have had such an impact on the AI breakthroughs of the last decade. Ioannis was a founding engineer at DeepMind, playing a key role in industry-defining moments from Deep Q Network and AlphaGo to Gemini. He recently left DeepMind with a banger announcement: raising $130M to build superintelligent autonomous systems with Reflection. We discussed his journey and the past and future of frontier AI. Nuff said. Let’s get to it.

Alex: You had worked as a founding engineer at DeepMind for over a decade before starting a company to build superhuman AI agents. If I'm not mistaken, DeepMind was the first company to embrace the term AGI, which has become somewhat bloated nowadays. What does AGI mean to you?

Ioannis: You're right. Starting in 2010, DeepMind was the first company whose goal was to build AGI. Artificial General Intelligence describes a system that can do everything a human can, given the same inputs and action space—a system with human-level intelligence that can perform meaningful knowledge work the same way a human can. You can delegate tasks to it and receive the results without human involvement.

However, as we got closer to AGI with technologies like AlphaGo or ChatGPT in the past years, we started to move the goalposts. The tendency now is to talk about AGI as a system that's an expert in every single field—the best that a human could be in all domains. Yet, this trait is highly unlikely to be found among humans. On the contrary, this behaviour is better represented by what we define as a "superintelligent" system or ASI, Artificial Superintelligence. Several companies have adopted this term recently. Unlike AGI, where the system achieves human-level intelligence, ASI is a system that is better than the best human expert.

Alex: I want to take you back to 2012. You have just graduated from university and joined a small startup developing computer programs to master video games. Was it like a handful of people back then? What led you to DeepMind?

Ioannis: I graduated from Aristotle University of Thessaloniki in 2011 and then pursued postgraduate studies in machine learning and AI at the University of Edinburgh. In 2012, I moved to London and joined DeepMind. We were 25 people then and joined as employee #6 in the Research team.

I wasn't sure anything groundbreaking would come out of it, but it was a very ambitious team. As a young engineer, it seemed like the coolest project, and I decided to embark on this journey instead of the consensus path of joining an established tech company where the scope of my role and potential impact would be limited. It was the right time to make a bold, risky move in my career. I remember thinking, "I'll just spend a couple of years doing really cool stuff, and then I'll get a real job". Well, things played out quite differently!

The company's mission was to "solve intelligence and use it to solve anything else". We started with video games like Breakout, Pong, and Space Invaders. Video games are self-contained worlds with clear rules and measurable outcomes, providing an ideal environment for training algorithms to learn, adapt, and strategise.

We went from systems that outscored professional video game players to winning the #1 ranked player in the world at the Chinese board game of Go—a game considered far too complex for machines, to predicting protein structures better than the most high-esteemed biologists and much more. It was a fascinating journey, to say the least!

Alex: You played a key role in all these industry-defining moments for AI. Your name is under the Deep Q Network, AlphaGo, AlphaZero, MuZero, and Gemini research papers. I don't want to sidetrack, so I'm leaving a link to the recent video I put together about DeepMind's story here for anyone who’d like to learn more. At what point did you feel that AI was reaching levels of superhuman intelligence?

Ioannis: I'd say there were three significant milestones. Deep Q Network (DQN) was the first aha moment. In 2013, we developed an algorithm that could master everything from shooters to car-racing games. It might be evident in hindsight, but we were still scratching our heads then. We weren't sure it was the right path. Seeing something working was an incredible moment and gave us more confidence that we would get intelligence out of these systems.

Then in 2016, AlphaGo won against Lee Sedol. Lee Sedol was one of the best human professional players in the game of Go. Three months before AlphaGo came out, people thought that beating a human professional in the game of Go was 10 years away. Three months later, we had this big game in Seoul that showed how these systems were already superintelligent. At that point, I could see we had achieved narrow superintelligence, in the sense that for something which requires a lot of intelligence, we had a system that does much better than humans. Of course, it had a narrow scope. It could only play Go, and then you had to take the algorithm and train it on chess or shogi. It wasn't the general system which could solve many different tasks.

Then ChatGPT came out, and also through my involvement with Gemini, I saw the generality of systems being able to complete a wide range of tasks. An AI program that knows where to start from, how to handle a task, and has some minimum level of intelligence to comprehend it and start trying to solve it. At that point, I felt we had all the ingredients to build superintelligence. This is what Reflection is all about.

Alex: We’ll get to Reflection in a minute, but before we do… What were the breakthrough technologies that enabled all this?

Ioannis: Several breakthroughs happened in the previous 10 years. Reinforcement learning and the ability to take these systems and optimise them using a reward model is undoubtedly one of them. However, several other technologies came together at the right time for general systems to emerge. The idea of next token prediction had been around for a long time in AI, but better deep learning architectures like the transformer, which came out in 2017, enabled the advent of LLMs. But ChatGPT wouldn't be possible unless we could amass internet-scale data and have the compute power to train these systems. All these couldn't have been possible 20 years ago.

Richard Sutton, in "The Bitter Lesson", argues that there are two methods we know will always lead to better performance in AI systems: learning and search. Today, they come together at a great scale, which can create general superintelligent systems. Big models, lots of data, better reinforcement learning methods on one side, and inference-time compute (leveraging more compute at inference time) with reasoning and thinking models like OpenAI o1 on the other.

Alex: You left Google DeepMind with a banger career announcement: Raising $130m to build superintelligent autonomous systems, starting with coding agents. I'd love to understand why you chose coding as a testbed and your vision with Reflection.

Ioannis: At Reflection, we want to build superintelligent autonomous systems. The way we think about it is that it's an AI model coupled to a computer, and you can delegate tasks to it. It's like giving AI access to your computer, and while you go on with your life, AI performs the tasks and gets you the results.

Then the question becomes: What's the best way for this intelligence to interact with the computer? Is it via the mouse movements? Via your keyboard? Is it the same way humans interact with computers? Humans have been shaped over thousands or hundreds of thousands of years of evolution. However, these systems were trained differently; they have different priors and biases. To them, coding comes more naturally. The natural way for AI models to interact with the computer is programmatically via code.

So, we concluded that the root node problem, which needs to be solved first to achieve superintelligence, is the ability of a system to reason and have a way to interact with the world that's well defined. And that's autonomous coding agency. The ability to try things, receive feedback, correct your assumptions based on this feedback, self-correction, in-context learning… All these capabilities are present in existing models, but we need to make them stronger to achieve superintelligence.

This is what we aim to do with Reflection, so join us if that vision resonates with you.

Alex: Playing devil's advocate here, several plausible bottlenecks could slow things down, e.g. limited compute, algorithmic progress becoming more challenging, not enough data. Going forward, what are the main challenges in building AI systems that are vastly smarter than humans and capable of novel, creative, complicated behaviour we couldn't even begin to understand?

Ioannis: We need to derisk several areas across data, compute and algorithms. There's a clear roadmap that shows us how to get there, which is something you can also sense in the industry. No one questions whether we'll get there. It's more of a matter of time. Some people say we'll get there in 2-3 years; others say it's more like 5-10 years. But there's an agreement among most experts that we have the ingredients, and it's a matter of putting them together in the smartest possible way to overcome current bottlenecks.

Yes, we should find ways to generate synthetic data, scale to even more compute, better algorithms, scale reinforcement learning further, train for agency, etc. But, I cannot foresee any significant blockers. It's work and the necessary steps that need to happen to get us to the next level of intelligent systems.

Alex: What are some AI research areas you are most excited about today and paying attention to?

Ioannis: I have been excited about reinforcement learning for a long time. Reinforcement learning is a paradigm of machine learning inspired by how humans and animals learn from rewards and punishments. The idea is to create systems that don't need to be explicitly programmed for every task. Instead, these systems can teach themselves by interacting with an environment.

After ChatGPT came out, up until last year, reinforcement learning had taken a backseat. The industry was all about pre-training and scaling data and compute. But I think now people have started to realise reinforcement learning is the way forward that will take us from something that just works to the superintelligence required to solve real-world problems. It's something we also saw with AlphaGo. Training big models was a big part of it, but being able to think more or less before taking an action was equally important.

Alex: You worked on Gemini before, and Reflection focuses heavily on language models. What’s the best way to approach benchmarks and evaluate the progress we make with LLMs?

Ioannis: We are getting to the point where the benchmarks we use to evaluate the performance of these models are saturated, and we have to come up with better benchmarks. But you know what? Academic benchmarks don't matter. What matters is solving problems for your customers and having an impact in the real world. This is why we have structured Reflection to be 50% product and 50% research. We are not aiming to build a big model that is good in benchmarks but does not provide value.

The other day, Satya Nadella, Microsoft's CEO, said something along the lines of "It doesn't make much sense to talk about AGI and superintelligence in an abstract way. We should measure its impact on the world economy. When the world economy starts growing by 10%, we know we are getting closer to AGI because we have unlocked something that wasn't there before." And I agree 100% with that. We can use benchmarks to keep track of our research progress, but what matters is the impact these systems have in the real world.

Alex: Before we wrap up… How would you advise a recent graduate who wants to get into frontier AI? Where would the 23-year-old Giannis finishing off his MSc degree start today?

Ioannis: There are many good opportunities for young people nowadays. I would strongly encourage people to join startups. When I graduated with my master's degree, I interviewed with many established tech companies like Microsoft. But I just decided to join a startup. When you are young, it makes sense to take more of these high-risk, high-reward bets and see where life takes you.

Especially among Greeks, there is this tendency to go for the safer bet. This is perhaps a cultural thing. Most families believe there is too much uncertainty in the country, so they teach their kids to go with the safer, low-risk path for everything they can control.

I recommend the opposite: When you are young, take as many risks in your career as possible. Join the craziest, most ambitious startups because this is where most breakthroughs happen. This is the best way to work on things that matter and have an impact. If you join an established tech company, you end up another cog in a big machinery. But if you enter a startup, you can immediately hit the ground running, as they say, and have an impact.

Alex: That’s a fantastic way to end this chat. Thank you!

Ioannis: Thanks, Alex.

Share

Jobs

A list of over 400 startup jobs in Greece from 100 companies hiring! Check it out here.

News

Runway (video AI) raised $300m Series D led by General Atlantic.

Galatea Bio (genomic research & clinical genetic testing) announced $25m.

Brainomix (AI imaging technology for healthcare) secured €16.7m Series C.

Wheelsys (SaaS for car rental & leasing) announced funding from Halcyon Equity Partners.

Wealthyhood (investing platform) raised €3.6m Seed.

Plugsecure (cybersecurity for electric vehicles) secured €720k led by Apeiron Ventures and Genesis VC.

Schema Labs (civic engagement) received a €100k grant from Helidoni Foundation.

Squaredev (AI custom solutions) was acquired (majority stake 51%) by QNR.

Resources

Sanne Goslinga and Marathon launching the 3rd edition of the Greek Tech Compensation Report. If you're a Greek tech company, you can participate here.

The distorted reality of startup success stories by Dimitris Glezos, founder of Transifex.

How Skroutz became Greece’s biggest online marketplace with George Hadjigeorgiou, co-founder & CEO of Skroutz.

The new role of PM in the age of AI by Babis Makrynikolas, SVP Product & Pricing at Blueground.

Nikos Drandakis, founder of Sync, on launching a car rentals marketplace.

Usage is the moat by Konstantine Buhler, Partner at Sequoia.

How a $1B SaaS turned user content into a growth engine by George Chasiotis, Managing Director at Minuttia.

Events

Join us today for Open Coffee Athens #122!

Talent Days 2025 by linq on Apr 5-6

Unleashing the Power of Gemma by GDG Cloud Thessaloniki on Apr 7

Bootstrapping a SaaS by foobar io on Apr 8

Harnessing GenAI on AWS Bedrock by AWS User Group Thessaloniki on Apr 8

ProductTank Athens meetup #7 on Apr 10

Silicon Valley Greek Biohub Kickoff Meetup on Apr 10

Practical AI Development with Python, Ollama, LangChain & RAG on Apr 10

Learn how to earn from gaming by ath3ns | community & web3 on Apr 10

AI Hackathon for Devs by Microsoft on May 3-4

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece—it helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X.

Thanks for reading,

Alex

• Why open source is the future of AI

• Are we getting AI-powered drugs soon?

• GPS and electronic warfare

If you’re not a subscriber, hit the button below and come aboard:

Subscribe now

From 0 to $20 million ARR

This was one of my favourite guest interviews to date. I had the pleasure to chat with Vangelis Mihalopoulos, CEO and co-founder of Yodeck, a global leader in digital signage technology used by 60,000 companies worldwide. We discussed the company’s journey since 2014:

• Starting from Greece in the middle of the economic crisis

• His biggest lessons in scaling from $0-$1m ARR

• Growing lean with less than $500k funding in the first years, getting rejected by VCs, and raising a large growth round in 2020

• Pros and cons of scaling a global industry leader from Greece

• Studying human psychology to influence his team and customers

• What he would have done differently about the Yodeck journey

Let’s get to it

Alex: Vangeli, super excited for this chat. You've built a global leader in your category from Greece in a very lean, sustainable way. How did you come up with the idea of Yodeck back in 2014? You also started in Greece in the middle of the economic crisis.

Vangelis: The seed was planted a few years earlier. I used to work for a software company, and at some point, we had a project to develop digital signage for pharmacies in Greece. Digital signage is using LED screens to showcase information, adverts, and multimedia content to public audiences. We had to plug every screen into a PC to serve the multimedia content, so the launch cost was significant.

Then, in 2012, Raspberry Pi came out. This credit card-sized minicomputer turns a monitor, TV, mouse, or keyboard into a full-fledged computing device. It was a breakthrough moment for digital signage, as suddenly, the hardware cost dropped significantly. You wouldn't need a large personal computer next to every screen. The launch of Raspberry Pi was certainly an aha moment for starting Yodeck.

Most importantly, founding a startup in Greece in 2014 was an act of “desperation”. Hard times force you to leave your comfort zone and take risks. We probably wouldn't have started Yodeck if we had a hefty salary hitting our bank accounts every month. It was an act of financial frustration. We had examples in our network, like Bugsense and Epignosis, that built global B2B SaaS products from Greece. If they could do it, why not us? That was the nudge we needed to follow in their footsteps.

We started with my co-founder Dimitris Tsingos, and two engineers joined us a few months later. It took us around 1.5 years to launch the first version of Yodeck to market — a digital signage online service coupled with a Raspberry Pi minicomputer to control any screen remotely through the Internet.

Alex: How did you get your first customers, and what are your biggest lessons in scaling from $0 to $1m in ARR?

Vangelis: We started with a self-serve model from day one. Our first customer was a diner in the middle of New York State who bought five screens, and we never had a single interaction. A community we tapped into to find early adopters was Raspberry Pi’s, and people looking for solutions built on top of such technology. They were all tech-savvy SMB owners, and most orders were a small handful of screens. While today, we even have orders for 1,000s screens!

If I can share a key lesson from the early days, it is never to sidetrack your business operations and strategy for one large customer or partner. During our first year of operations, we had one big customer that could potentially lead to an acquisition. To understand the scale of the opportunity at that moment, we had sold around 700 active screens in total (all SMB inbound leads) and that customer wanted to buy 55,000. We diverted our roadmap to meet their demands and spent a lot of company resources for over six months. That deal eventually fell through. Looking back, it was expected because our business was not built for that type of enterprise sales. It taught us how we should stick to our go-to-market motion and work for the long run.

Alex: If I'm not mistaken, you raised less than $500k in total in 2014 and 2018 before a growth equity round in 2020. All these years, you stayed lean and grew sustainably. Did you have options to raise more external capital in the first years and decided against it? Looking back, was this the right move?

Vangelis: We raised our first money in 2014 from Startech Ventures, the venture-building company Dimitris, my co-founder, has been running since 2012. Then, in 2016, less than a year after our first customer order, we started reaching out to investors to raise more capital. We had a few hundred customers but were not cashflow positive yet. We entered conversations with three VCs in Greece and, in the end, were rejected by all of them. A common feedback we got was that digital signage was not a "hot" tech industry. That hasn't changed since. We are talking about a 30-year-old industry.

However, investors’ appetite shifted after Yodeck reached a critical revenue size. We were no longer regarded as a "high-risk" early-stage startup in an unsexy field, but they evaluated us based on traction, revenue, and growth rate. Not raising more money in the early days led us to grow organically and build a solid foundation for the years to come. If you ask me, running a company like this also suits my personality type. I'm not the risky, growth-at-all-costs entrepreneur. Connecting the dots now, sustainable growth was probably the right path for Yodeck.

Alex: What changed in 2020? Why was then the right time to raise a larger growth round?

Vangelis: It was the right time for a couple of reasons:

• We wanted to be more aggressive in spending. We had been running the company in a break-even mode for years and didn't have the resources to take more risks. Raising money would enable us to accelerate hiring and go after several opportunities we thought would establish Yodeck as a global industry leader.

• COVID had just hit, and for the first time, we saw our annual recurring revenue (ARR) dropping 3% in a month, whereas until then, we were used to a 10% growth every month. That made us seriously consider raising additional capital for an extra cushion, as we didn't know how long the COVID effect would last. Luckily, the demand ramped up quickly, so cushion was never needed.

Alex: I did my stalking beforehand, and in one of your recent interviews, you said the company recently crossed $20m ARR with an impressive customer clientele. Where is Yodeck today?

Vangelis: We surpassed $20m ARR three months ago with a really strong YoY growth rate. We serve close to 60,000 businesses around the world. Half are paid users, and the rest use our freemium product. The majority are SMBs, so we have a large revenue distribution without overly relying on large customers. Our team is currently 140 members strong, and we plan to hire another 50 people this year. If you want to join us on this journey, check out our job openings.

Alex: You have a lean sales team for a $20m ARR company. What are the key pillars of your growth engine?

Vangelis: Until 2023, we focused exclusively on inbound sales and a self-serve model, where customers used our product without interacting with our team. We had a small sales team that contacted customers only if they submitted a form requesting to contact us.

However, inbound sales can only take you so far, and historically, companies typically face a plateau of around $40m-$60m in revenue. Therefore, a year ago, we started increasing our sales capacity to identify opportunities to expand existing customer accounts. We are thinking of outbound sales as more of a strategic channel. We approach accounts where a handful of sites already use our product and try to expand, perhaps at a global scale.

At the same time, we are building a branding team to drive more leads at the top of the funnel and a team to support our customers via phone calls. We ran our first customer summits (in the US and Greece) last year and are preparing more events in 2025.

Alex: Correct me if I'm wrong, I think most of the Yodeck team is based in Greece. What have been the main pros and cons of scaling a global industry leader from Greece?

Vangelis: Indeed, most of our functions are based in Greece, outside a small office in Cyprus (soon to reach 12 employees), and 2 executives based in the UK. We have great talent in Greece, and with the proper guidance and training, you can build a successful global company from here.

Compared to the US or most tech hubs in Europe, you can build a team you can count on staying for the long run, and that's very important because of the tacit knowledge that your team develops. Since the early days at Yodeck, only 11 employees have resigned and decided to leave the company, with just one in 2024. This highlights the great work environment we have developed, being recognised in the Top 10 of Best Place to Work for 3 consecutive years. Perhaps the global job market also plays a role, or it might be just a cultural thing. At the same time, the cost structure in Greece is more efficient compared to the US or other tech hubs, which is our “unfair” advantage against others and allows us to remain price-affordable and competitive globally.

The decision is also emotional. Having almost the entire team based in Greece is like giving back to our country. We were born and raised here; this is where we live. We want to support the local community and tech industry.

On the other hand, very few people in Greece have experience selling technology products to the global market or witnessed first-hand companies scaling to tens of millions of revenue. Therefore, we train and mentor our employees to keep up with the company's ambitions. We have been doing this for years, and we aim to allocate more time and resources to it. People do their best when they work for someone they admire and appreciate, as they get inspired and feel they can learn from them. This is why we train our best individual contributors to become people managers. These are capable and skilled people who can grow into great leaders with the proper guidance and due time.

Alex: You have been leading Yodeck for over 10 years now. What's an example where you had to grow as a leader, and what was the impact on the company?

Vangelis: Where should I start?? A lot has changed in the 12 years of building Yodeck. Today, I act mainly as the "CPPO", the company's Chief Product & People Officer. The psychology of our people (how they think and make decisions) and the psychology of our buyers (how they scan the market, make purchasing decisions and actually use the product) are tremendously vital for our success. Before starting Yodeck, I would never have imagined that I would spend so much time thinking about human psychology. Understanding what drives human behaviour and being authentic have helped me win people's trust all these years.

Alex: Knowing all you do today, what would he have done differently about the Yodeck journey?

Vangelis: I would have doubled down much earlier on learning and development. I would have worked more actively preparing the ground for our best people to become people managers and rise in the managerial ladder. We saw that the organisation was growing fast, and we would need to build our leadership team to support the next growth stage. Yet, we allocated proper resources only when necessary instead of planning a lot in advance. That’s probably the main thing I would change if I could go back in time.

Alex: Thanks a ton, Vangeli, that was fantastic!

Vangelis: I hope that was insightful.

Share

Jobs

A list of over 400 startup jobs in Greece from 100 companies hiring! Check it out here.

News

ReflectionAI (superintelligent autonomous systems) launched with $130m in funding from Sequoia, CRV, and Lightspeed. Stay tuned as I have the founder on Startup Pirate soon to share his journey from founding engineer at DeepMind to building ReflectionAI.

Axelera AI (hardware & software for edge AI) announced a €61.6m raise.

TurinTech (AI-generated code) announced $20m in funding.

Lastmily (last-mile delivery software) raised €1m from Unifund.

ORamaVR (VR medical training) raised funding from Evercurious.

Expand network (API connection for DeFi) was acquired by Blockdaemon.

Nvidia set up shop in Greece, already having a small photonics R&D team in the country.

Resources

Reindustrialisation led by defense technology by Panos Papadopoulos, Partner at Marathon VC.

The journey of Taxibeat from 0 to $43m exit with Nikos Drandakis, founder of Taxibeat.

Building a new type of defense prime in Europe with Dimitrios Kottas, founder & CEO of Delian Alliance Industries.

From a passion project to a cybersecurity industry leader with Haris Pylarinos, founder & CEO of Hack The Box.

Choosing the right startup exit timing by Dimitris Glezos, founder of Transifex.

Reviving "Made in EU" by The Greek Analyst.

The real role of a Data Engineer with Konstantinos Siaterlis, Senior Cloud Engineering Tech Lead at Mondelēz.

Web & google analytics, data stacks, and more with Charis Maragkakis, BI Analyst at Welcome Pickups.

Productivity tips & tricks by Christos Chatzis, Sr Technology Manager at Kaizen Gaming.

Events

Open Coffee Athens #122, join us on Apr 4!

DevOps Strategies for Power BI by Athens Power BI User Group on Mar 21

Kubernetes Athens vol26 on Mar 21

SuiSocial - Athens Suihub Community Meetup on Mar 24

AbZorba Games: From inception to scaling, then to exit and beyond by Scaleup Greece Meetup on Mar 27

GraphQL: One API to rule them all by Thessaloniki not-only Java Meetup on Mar 27

The robots are coming: special topic Superintelligence on Mar 29

Ask the CTO by Leadership Conversations in Tech on Apr 3

Beyond Tomorrow Startup Village | Beyond Expo on Apr 4-6

Entrepreneurship and ML by Patras Tech Talk on Apr 8

Take Back the City # 6 - SKG by Astylab on Apr 10

Devoxx Greece on Apr 10-12

Panathēnea Festival on May 7-9

Greeks in AI 2025 on Jul 19-20

Subscribe now

That’s all for this week. Tap the heart ❤️ below if you liked this piece—it helps me understand which themes you like best and what I should do more.

Find me on LinkedIn or X.

Thanks for reading,

Alex