George Hadjivarnava: Founding Foody, and building AI at Terra

That is good, and it's good that it happens. At the same time you need to help startups in Cyprus actually monetize and get money, not just proof of concept and nothing. You don't have to consider every startup as a unicorn to build an ecosystem, an infrastructure, supporting services, and customers for new startups that can actually start here and go global. You need to have this ecosystem around you. From basic things like shipping in Cyprus and things that support downstream integrations with the tax office and integrations with banks. There's still no open banking API in Cyprus. That's crazy. I can see my Bank of Cyprus balance on Revolut now, but I cannot easily build an app that does that in Cyprus as a layer.

Yeah, clear monopoly. You have the APIs, I can see that on Revolut as well. It's two banks, it's super easy to build the APIs, clear monopolies, and then you don't see actual apps being built on top of these.

Exactly, because it's very hard to actually get into the ecosystem and establish that relationship with the banking infrastructure. Plus some other stuff. When we were doing Foody, we wanted to automate all the payouts to vendors.

To a restaurant.

And there was no way of doing it without human interaction. Even on the API side with the banks, every three hours the token would expire and someone had to be there to re-authenticate the process. It does save you a lot of time, but it's not fully automated. So you need that intervention there.

Before we get into Foody, you spoke about the ecosystem of startups. Someone from the outside would probably not understand why this is important.

It's important because if you look at Silicon Valley, you have startups and almost every startup is a customer of another startup. That's how you build an ecosystem. You build supporting services, and you can start investing in tech and software, because if you have startups as customers you can provide value added services, essentially software as a service for those customers. You build an ecosystem that is localized at the beginning, like the ecosystem of Cyprus. There are a couple of startups now that do accounting, integration, and tax submission, trying to help businesses automate these processes and provide software as a service specifically for Cyprus. If you do that, then you allow other startups to come in and not spend time and resources on this part of the business, and focus on what they do best. If you start to do this as a network effect and many startups start to appear in Cyprus and they actually monetize, so they have money to spend, and it's not just funded, it could be self-funded, then you can build startups on top of them and serve them. You have this ecosystem of infrastructure and it scales up, and you have more layers that you can build on top. We saw how it happened from 2008 onwards. Then you have the Ukraine war, you have other conflicts, you have people relocating their whole teams to Cyprus. I don't know what kind of company, but you have jet planes, they brought their whole team to Cyprus. You see big startups advertising for positions, hybrid, remote, or local in Cyprus. So you have big teams, and these people have needs. You can go knock on their door and make direct sales with them, not a cold call. That's how you build the ecosystem, and it's a two-way relationship. A big company comes here, you try to sell to them, but they sell something as well. In the world we live in, where the market is global and a company can sell to anyone in the world, the in-person relationship sometimes is more beneficial. Nobody can make a sale in a single conversation, but if I go and knock on their door now to make a sale, they might make a sale as well, or not, but the relationship we build is very important. Because there are locals in Cyprus now, there is this community effect, and that is probably the best way to do business.

Are you actually optimistic from what you've seen the last ten years, or pessimistic?

From what I've seen the last ten years I'm probably pessimistic. From what I've seen the past one and a half years I'm turning to be optimistic.

What's the one and a half years?

The one and a half years has nothing to do with politics or government. I just met a lot of people that are here and they like it. They want to make something out of it. As time goes by, the generation gap increases, and you see new generations that have a three year gap. I talk to a lot of students, high school students, and university students. I was always a fan of that. You see people in high school understanding already that the current education track that one will follow is not so viable anymore. So you start to see people investing in themselves earlier. When you do interviews for software engineers, you get 17-year-olds who are really confident they can compete with senior people who are 35, and they just did that over a period of two years. And you see those people in Cyprus. You see people who are 30 but never went to college and made something out of themselves. They are consultants, or they did contract projects, and they are here now working for people all over the world and for Cyprus. You see that they start to build communities. If you see communities growing, and not spam communities, if you see people helping each other and wanting to help each other, you definitely feel more confidence that this thing will actually take off.

We're speaking about the Cyprus ecosystem here.

Yeah, and that is positive now. You see communities building around it.

I would counter your argument, to be honest. I agree with you partially, but it seems to me that we are focusing a lot on the optics. By saying that, I kind of see a repetition of what you said earlier about the accelerators. You hear accelerators, and then you hear funds, and it seems to me like some sort of an effort from the government to say, we start from the money and the picture and the optics, and us trying to convey good stories of what exists in Cyprus. I think the opposite approach should be applied. One should say, look, we didn't have a lot of successes. The only company I know of outside of all the Forexes is Foody. When I was growing up, the only story I was hearing was Foody, and I was using it and seeing the approach. It was the only case where you would say it's a big company, big culture, big acquisition. But the government comes and makes all these claims that we are investing in diverse founders, we are giving money here, there is this VC, and we're trying to portray that there are so many successful stories. The reality is that there are no successful stories. There is one successful story, and one should embrace that and say, look, we don't have technology, but we should bring the technologists, enable them to change what we have. Let's not focus on real estate or all the incentives they gave people to get a passport. You could do exactly the opposite and say, whatever the incentive is, I'm going to enable entrepreneurship, I'm going to enable startups, and I'm going to enable technologies. Once you do that, you start going to universities and you invest in people and in technologies. In the Valley, if you go to a batch today, you see 18-year-old kids reinventing the world. They raise tens of millions of dollars and go out to the world and start building. Some of them become Airbnb, some become Stripe. If you are 18 years old, the person who is 40, 50, 60, who built their whole experience, is actually irrelevant. If you compare with AI, you can build a whole new type of business that didn't exist in the past, and it's a level playing field. So I would avoid optics, and say, how do we enable these people who are going to build new technologies and new change, and stop all of the optics game. And also enable the people who did this in the past.

I agree with you 100% there, and I'm not a fan of optics. I am pragmatic though, and I understand that politicians will always want optics, because that's the world we live in, that's how votes are acquired. I don't want to argue about that. What I can say though is that people try to do things, and I focus a lot on the community. I repeatedly say this: they are our communities, and at one point they will influence the government. The government should do it in advance, the government should realize it and do it differently. 100% I'm with you. Even with the real estate and the passport scheme, at one point there was a fund where a percentage out of every passport investment would go to innovative startups. But if the investor didn't know any startup, it would just go into a black hole.

You remember the kind of forms that you need to prove that you are an innovative startup and all of that.

Yeah. You see even different funds that come from the EU. You're saying at YC 18-year-olds get 10, 15 million in funding. We're talking Cyprus in 2013, I think the innovative startup fund was 15 million euros. I think they managed to give out 1.2 million over a period of three years, and then they took the fund away. These games are crazy to think about. Besides the optics, I think in the backbone some people do operate in a way that is favorable, but nowhere near what you're saying.

It's not a confidence question, because the confidence is there. There are incredible people who can build incredible things, but one should be able to give opportunities. We need to move away from these approaches where the accountant is going to teach you business, the lawyer is going to teach you business. Why don't they get George to advise startups if they want someone to build startups? That's a key core question for me.

That's a problem with the optics though. In the panels you bring CEOs of traditional banks, people from the Big Four. I'm not saying they're not competent to give advice, but they're not the people you want giving advice. You want people who can give raw advice and don't care about the optics of the advice they give. They speak from experience, and they can provide advice under a certain context, because not everything is the same. You say the playing field is level now, but it's not at the individual level of every startup. Everyone is solving a different problem, and although you can use the same tools and the same approach, getting advice from someone fully in your context is probably the best advice you can get. Everyone can pick up a book, everyone can read the best blogs, everyone can listen to the best podcasts. It's all about bringing it to context. Besides the optics of the government and how these things are approached, I would like to see more opportunities and a push for people to actually be able to innovate in Cyprus, and say, okay, I'm 17, 18 years old, I have the support and the confidence to go out and do this, and understand that I'm probably going to fail, but it's going to be one of the best experiences I have. It's going to be bigger than university, bigger than anything else. I'm going to suffer, I'm not going to have the pay that all my friends have, I'm going to be grinding on the weekends, and I want to make this happen. That push is what is missing. Everyone sees the beauty of the startup because there are some success stories nobody knows about, some self-funded success stories, and big companies that are traded publicly that are Cypriot-owned, and nobody knows about their owners. That's not the issue. It's not that we don't have success. We don't have the push from society to actually go and try these things. We are a very fresh democracy in Cyprus. 20 years ago our parents had the mentality that you need to get a job at the bank or at the government to have a secure income, and that's it. So the culture is like that. I remember I was consulting with a US company that was opening in Europe, and I was talking to the owners while listening to a podcast. The difference is that in the US the dream is to be a family man and have your own business. In Cyprus it's to be a family man and have a good job, or actually a house. So you can see the difference in how society and education prepare you to become a successful entrepreneur. This is deeply rooted in the culture. So I'm not going to focus on optics by the government, I'm going to focus on this side of things, not even on the infrastructure. I want to put this in the minds of young people, that you can do it from here and we will support you in XYZ ways. Even if you fail, it's actually making you more prepared to do the next thing, whatever the path is.

Yeah, it's always worth understanding the question of what was the culture.

Basically just to make rent, that was it. And minus the spending. It was not what we have now in London, but yeah. And then I realised the network effect that Imperial had. I could meet people. I finished university, and the fact that it was 2009, the Goldman Sachs crisis reached the UK, Europe, and that's how I decided to come back to Cyprus. My plan would be either to go into one of the investment banks in London and start working there, or I got a job in the local Telecommunications Authorities in Cyprus.

Why is everyone leaving Imperial, even today, wanting to go to some sort of investment bank?

That's part of the Imperial process. If you did a sandwich course with engineering, you did the summer working period at one of those banks and they hired the people from there. That's how it happened. Plus, it was the best way to get a solid finance package right out of university. It was the sole career path. Most people from computer science courses that didn't do any deeper tech ended up in investment banks. That was the goal too. So I came to Cyprus working for this secure job that we had before, as part of the culture, but it was not enough for me. I couldn't be in a place where I just look at the clock to finish what I do, because I couldn't innovate there. So with friends I always started to do different things, from the online ticket sales website to listing marketplaces, and that's how I met the co-founders at Foody through At Your Service.

Did you meet Michael or Argyaridis first?

Argyaridis back then. With my partner we were doing At Your Service 2.0. It was a small software firm we had called Brain Bile at the time, and that's how we met Argyaridis. And at one point, one thing led to another, Argyaridis told me to join. The first thing to do was monetize a platform, which was a headache.

How did he convince you, actually?

Many conversations. It was actually hard for me, because I think I had my first child back then. So all the vetted, stay-secure job culture that I had, I had an eternal battle. But then I knew that I was going to get bored very easily, so I needed to make the step, and that's how I made the step. Probably one of the best decisions of my life back then. Self-funded, no investments, nothing. And two years in, we started building Foody as a pivot. That's when we had to get serious and actually start building, applying things that we learned from the YC podcasts, the blogs, the books, the US success stories. Other Cypriot players that were actually in companies, either as co-founders or founders or were just working, so we could apply things and experiment. And that's how the whole thing spun off.

What technologies existed then and what exactly didn't?

Man, I was a PHP guy back then. So learning empirically, computer science knowledge from university comes at a much bigger scale. When you start something, you don't think about that. You just want to make the product and iterate, see what happens, validate the business first. The first version of Foody, we were making calls to the restaurants ourselves. You would be ordering online, but we would pick up the phone. We had our ordering system integrated with Twilio to call us whenever there was an order.

So basically a landing page, an order comes in, and you place the order in the code.

Yeah. At the beginning we would get an email, personal emails, and we would make the order. Then we made the first Excel-style back office, and we continued to build on that. No online payments casually at the beginning, and three machines. It was huge already.

No online payments. Were you dealing with a JCC gateway stuff early on or not?

Yes. Actually, fun fact, at one point when Foody picked up, we got the double primary key on their main database from the traffic. But we worked with JCC a lot, so that's what I was talking about before. You don't need the government. At one point, if you get big enough, you influence things. You change technology, you change culture, you work with the infrastructure companies and you make things. This is why it's important for people to start building with the hope of getting big, to influence.

Mano, on that though, my counter would be this was also the experience of Stripe. But then it's the ambition that led Stripe to do all of these massive changes, because for them it was also the Uber story and the DoorDash story early on. And then my segue here is the vision, right? You can think of the JCC part as they were building useful things, but they just narrowed it down into Cyprus, which could be so much bigger if it was outside of Cyprus.

A hundred percent. It doesn't make you dominate a local market though, because it is.

No, absolutely not. It's exactly that. You enable the local market, but then you start thinking, well, maybe other people have this exact same problem in Greece or in the UK, because Stripe is just a very recent story. It's not like 50 years old.

That's true. We are in Europe travel still, because you talk to everyone in the US that comes to do business in Europe and they tell you about the European compliance and how it doesn't allow people to operate here. It's not the United States, it's individual countries. They have a common governing body that is the EU, but the taxation system, the EU Inc. is just getting traction this year after the 10-year effort. So you could at one point register an EU company and have a uniform taxation and operational model across all European countries. So it's very hard. You see most startups are started in Cyprus and establish themselves in the local market, then they go to Greece, then they go to the Balkans, and that's it. The traction cannot expand beyond that.

Which I see as a bad idea, to be honest. There is always this step system in Cyprus that we think of: let's start in Cyprus, and then we're going to expand in Greece, and all of these are small markets. The level of difficulty is always about the market you attack, and it doesn't mean that building a bigger business is actually more difficult or more complex. It might be the opposite sometimes.

Sometimes it's exactly the opposite. To make a good business in a small market like Cyprus, the approach is totally different, because it's a market of supply and demand, especially if you are B2B2C and you have the chicken and egg problem. You need to focus on both vendors and customers and end users at the same time. And you need to build very strong relationships, because word of mouth is stronger than ever in small markets. So the methodologies you apply to establish yourself, provide good experience, have good relationships with everyone, don't apply at scale if you go on a global market where it's faceless. That's why it's a problem when we start from Cyprus and we go to Greece. Greece might not be the biggest market, but it's incomparable to Cyprus. You go from one million to ten million. Every time you jump an order of magnitude, add another zero, the dynamics change completely, from hiring to vendors to users to everything. The approach is totally different. Then you go to the hundred million user market, which is even more difficult to have direct relationships. I don't know if you hear advice that as a startup you shouldn't have too big customers and not care about the rest. But at the same time, in order to build something big like Stripe, what they do with Uber and DoorDash, you need those customers to build for them, help them, and they help you at the same time, not just as social validation that you're working with this big company. You see the variance and all the nuances that come when you have to serve a customer as big as them. Things appear that you don't expect to have. But they help you become better, innovate, and provide smarter solutions to actually make those contracts work. So this, from Cyprus, is missing a lot. And I want to go back to real estate. Real estate is such a big part of our GDP, but we are building the same way that we built 50 years ago. For big projects, if you go to Limassol and see the big towers, companies from abroad come and do this. For me, that's the biggest shame of all, because it might bring people here as well that they spend, but you own it.

Yeah, it's money. It goes back to the diversity idea, which is we want to diversify Cyprus outside of the things it's good at. And I think one should do the opposite, right? You shouldn't actually say I want to be good at everything to diversify the government. You should say, well, there is product market fit in tourism. There is product market fit in real estate. There is product market fit in these five things, and we should be the best in the world at them, because the market allows you to do that. Even in the case with the occupation, you know that Turkey is here, they occupy half of the country. You can actually build military, you can build systems. Maybe Israel is a good example of that: there is a need, and if there is a need there is product market fit. If there is product market fit, you're going to actually be building for these types of markets. So the diversification argument is completely the opposite of what it should be.

Exactly, and it's not just a matter of becoming the best or not, it's a matter of making the infrastructure and innovating on this so that you can sell it abroad. On every market that is in a bubble, usually you need to have compliance and audits and everything, and you need to enforce them. One thing that I would argue with someone in the government relationship in Cyprus is to say, okay, we have all this real estate happening, we have an issue with auditing, we have an issue with compliance. Allow people to build businesses on top of that. Give contracts to smaller teams so that they can start to help their ecosystem with managing this process. Give it to the private sector to actually do it for you. This way you provide the framework for new startup streams. Because if they do it well here, where you have a big market on real estate and a lot of real estate, and money coming from all over the world, then yes, that team can go and sell abroad to anyone in the world, because they did that very well.

Well, it's basically the Singapore story, right?

Exactly.

They decided one day that look, we're going to export talent, we're going to export banking, and we're going to be the best in the world at it. For fifty years, they really achieved that. From a third world country to one of the richest countries in the world.

And it's a very similar story to Cyprus. British Empire, same situation that we have with Turkey. But in the end it's okay. We're going to abolish everything, language, religion. We're going to focus here and become the biggest financial market in the world.

And they opened it like a startup.

Exactly. And they achieved it.

We spoke a bit about the technologies that you built in the past at Foody, and we spoke a bit about payments. Walk me through the transitions in technologies that you worked on. You also stayed at Liberty Hero for a number of years too. What's your view on the technology shift over the last 10 years to what it is today?

Did you guys build POS systems, for example?

We integrated with existing POS systems. There are POS systems from student projects that happened at the JCCI level and they still work. And you have companies that do partnerships with the big software houses from abroad, popular tools or local tools that are expanded vertically in the sector, so you have to integrate with them to minimise the overhead on the restaurant level to actually process an order. So with the acquisition you had to integrate with different systems, and they apply different modes of communication. At one point we needed to migrate a big part of the platform to central services while maintaining it, because people continued to eat, they continued to order groceries, and Foody was still growing. Did the technologies at hand really change? Not entirely, but the methodology had to change. When you're in a startup at the beginning, it's all about getting the traction and reiterating a lot and building trust at a different level. When you get to the midpoint, you are established, people allow you to make mistakes because you already show the value. But then you need to start showing that you are an adult now and you are going to be reliable and resilient. So the methodology of doing things changes. You cannot have the YOLO ship to production thing on a Friday night just because you are a startup and you will be excused. So the whole approach changes. And transitioning to what we have now, with agentic development and all the tooling and how fast you can make an MVP, the fundamentals of having a resilient, maintainable system that you can build features on, and not be locked into a vendor, still apply. So the methodology, everything: even if you take agentic development, you can one-shot an app that will do anything you want, but you cannot one-shot an app that will scale to a hundred million users. The same methodology applies. The groundwork is done by the agent, but the ideas, the data points, where you want to focus, and what actually happens in the code still need to be part of the product development life cycle.

And do you think that infrastructure now is even more important?

For sure.

And do you see this changing? Do you see agents actually figuring this out?

You see agents figuring this out, but there is a gap. There is a gap because everything is made for people. There are a lot of things, and I don't want to go into semantics like skills or aggregators or APIs or documentation. You see a shift, but the experience is not agent-native. It is not meant for agents. So you don't have this layer where the communication is for an agent to figure things out. If you strike the agent in natural language, the agent needs to translate that into their context and then into, I don't know, the documentation of an API, and then do their work. As a human, essentially, how a human will develop, will integrate, get the results, come back to you, and communicate the outcome in natural language. You have all the steps in the middle that are just data transformation essentially. And every time you transform the data, you lose context. So there needs to be a layer that bridges the gap.

So why do you lose context every time you translate?

I don't want to go into specifics of how LLMs work, but it's a simulation of a brain, and it's a simulation of a brain in a context. We cannot see ourselves as humans. Like, we talk now about software development and agent communication, but if we switch to football, it's going to take some time for us to adjust, and we cannot carry all that bandwidth together. The same happens with the agent. The more shifts you make in context, the more context you lose. And if it's just a way of transforming information from one layer to another, you lose context without meaning to.

But I think that's maybe the current big thing, context engineering, which is figuring out how to enable the agents to understand the information and operate on that information.

Exactly, yes. Don't get me wrong, that's what I'm talking about. What I'm trying to say is that it shouldn't be done in the traditional ways of communication and documentation about systems. It should be a different layer, because at the end of the day the user query comes in natural language. It's not just the exact query the user says. Someone has to identify the intent behind the query. So that's where agents help us even more. If we help agents understand that intent better in the context, and build layers that are not just APIs, they understand the context of the query. They actually respond, and if they don't understand it correctly, they respond in a way that they can understand it, and the communication happens there, not on transforming an OpenAPI spec to an SDK and a code base. That's not where this reality is.

Yeah, exactly.

So where, obviously you can't call it a transition, or there might be some more interesting word, but Terra operates with so many developers, and I always get these types of questions like, which type of app and which type of developer? But the reality is they're just so varied. There are people building wearables. There are people behind hospitals. There are people behind AIs, and they're such different use cases. But there is always this layer and this infrastructure that you can keep building on. And all of this time we've been building this technology that started with the siloed information: there is siloed information about your health, and it sits in a hospital, it sits in a device, it sits in a health app, and it's 300,000 new apps that come out every year, right? It's crazy. And all of that information stays there. So initially that's the problem we were solving. Then it transitioned to a problem of, look, there is all of this siloed information in so many different places, and first you build the connectivity, then second you enable developers to access it in an easier format, in a normalized format, and so on. But then there are more ways to be useful to people. So it's, can I enable data access? Can I enable blood ordering, and so on? There are just so many more services that you can add on top. And then obviously all of these companies out there, there is Oura and there is Whoop, right, that came out and said we're going to offer new blood services and so on. So it's always a question of how can we be more useful over time. But then the transition I'm speaking about, it's been changing for the last maybe year in the way that all of the AIs start using Terra, and then the question changes from, well, how do I serve a developer, to how do I serve these AIs? And AIs are so different in the way that they consume information, and working with the big labs allows you to see behind the curtain, to see where things are moving. So how do you see that? What are the new learnings that you get that people might want to hear about?

Okay, yeah, I see where you're going with this. If we look at AI and how it started, and how automation started with the LLMs, we went beyond the ChatGPT ask-about-anything mode and we started building things. The first thing that came out was MCPs. Then MCPs are not so popular anymore and now you have skills. You already have some communication protocols, let's say, that are not protocols in the traditional sense that are very strict and documented. They are ways of working with the model. These are more focused on human-to-model interaction. But you don't have model-to-system interaction. There is the MCP, there is a skill that maybe describes how to access this API, but that API is not meant to communicate with the agent, it is not written in a way that it can communicate with the agent. So I think the next layer is not building APIs and normalising data, the next layer is actually helping those agents communicate with systems without losing context. I think this is where the big play is now, and this is the next layer that will come to every industry, from payments to health data or wearable data, or someone tracking their path at the individual level on anything they do. So that layer is what is missing now between the agent and the systems, because between the end user and the model, it is pretty much done since five years ago. The first version of ChatGPT actually established that. So that's not the problem anymore. The problem is how the agent on its own, in the background, without having to be prompted every time, communicates with the systems, without having to go through documentation that was built for humans and developers in the past. So there needs to be a layer where you transfer the context as best as possible and you have an interactive communication. It's not one query and one answer anymore. There needs to be interaction that happens and not just a direct response. It could be over a period of time. We could have an agent working on something with a system, and three or four days into the process they need to query back the user to get some more information, or they need to create another system, or get an approval. These layers don't exist now. And that's where the next shift is, I think.

It sounds like we're describing the creative process. It's like if I start writing something down, I'll go for a walk or speak with someone, think about this idea, write something down, then go back, start erasing this, and it's just going to take me ten days to compile this idea onto a piece of paper. But it sounds like this is where it's going, isn't it?

Exactly, yes. We always think about AI and the agents as something that can do something fast. This is the showcase, but it's the optics we talked about before. The real value is actually being able to leverage them to maximise the value you get, to maximise the autonomy of the agent, but be there when you are needed. The human-in-the-loop part will always be there. So that's where we need to optimise, and that's how systems work anyway. You don't want an agent to do one soft task. You want an agent to actually be an agent and do things for you. So there needs to be a way that you compile that idea, as you said, you build on top of it, and then you let it go again, it works, and then it comes back to you. For us to be able to achieve this, we need to have that layer of communication so it can work autonomously, it can gather information, but then it comes back to me. Because at the end of the day, I'm the one who's driving the system. This is my argument against the standard talks that AI will take everyone's job. It may take everyone's job at one point, but the level that everyone will be able to operate at and the work that everyone will be able to do is completely different.

It kind of sounds like the agents are today still being used like Google by the majority of people, which is search, right? Whereas the majority of the work that we're doing, and where this is going, is how can I have a system that I give time to and go for a specific outcome, right? If I parallelize this with a really bad parallel, it's something like a football team. You want to achieve a goal, which is getting the league. Let's not speak about that. You have someone that does all of the transfers for that team, and someone that trains that team, and it goes day to day playing the games and so on, and it takes a year's time to actually complete that goal. But it's that process of doing all of the right things to achieve that goal. And in health, what I'm really excited about, where this could go is: if you decide that, look, I live in Cyprus, but I one day want to be an Olympic athlete, or, longevity is trendy now, so maybe I want to live until 120 years old, or I want to be the best student and end up in Imperial College and get the best grades. You provide this system with information and you say, here's my DNA, here's my blood, here's my time series information coming from wearables and all of these health apps, and then this system constantly sees the world and says, hey, I saw that you ate this today and I saw that you're training that way, I think you should go and do these five things, and then it constantly looks at your information and you end up as close as possible to that goal, instead of listening to some sort of accident that happens one day, and you just learn that you have this deadly disease out of the blue that you never really considered. And going back to Imperial and engineering, it's more of the feedback loop idea: there are inputs, outputs, feedback, and you allow yourself to become this engineering system that constantly optimizes for yourself and for your best outcome, instead of just learning from a static sample one day that you could always have seen and investigated prior.

It's exactly that, and it doesn't have to be just target-based, it could just be monitoring-based. Having a target definitely makes you more motivated to have this system helping you. But we always talk about the advantages of AI being how our health will be in a much better place and we can monitor everything and have the AI help us in this. It's not just that though. Nobody has access to continuous monitoring, with someone looking at everything that happens to them and being fully dedicated to them 24-7. A doctor cannot do that for you. Even if you are a billionaire and you have a team of doctors with you every day, they still go to sleep, they still have to communicate, they still have to share a way of working. So AI can help you bring all these things together and start looking at things progressively and building everything on top as you go along, because the methodologies change, people change, habits change, even the blood sample you get changes, and the way you get the blood sample changes. So in order to keep track of everything and keep it together and get the feedback loop with you or with anyone else in the meantime, and build on top, nobody can do it better. You need someone that will be there 24-7 and respond to any input and ask the question that it needs to ask to get the best possible outcome.

And that's the role we are opening now.

To be a bit practical, Terra roughly receives twenty thousand applications every year, now twenty to thirty thousand. And the problem is how do you recognise if Nikola Tesla applied? It's an almost impossible problem, because if you consider the most successful companies in the last ten years, it is some sort of CRM in sales or CRM in hiring. And it's not really solving the hiring problem. It is a good presentation of the people who come into your pipeline, which, if you think about it, in the AI era is so dumb. Whereas it used to take roughly three weeks to do an interview process with a candidate. Now we just built this tooling that first filters for the information I need. You can provide it to the AI and say, here's the type of person I'm looking for, here are the traits, here's everything you want to understand about that person. And the application process goes from answering my questions and sending my CV, to a video interview and a coding interview, and all of these stages are actually evaluated by an AI. The majority of it, like ninety percent, is just done by an AI. So now, instead of having three weeks, you can compact these and say the last interview stage is going to be with three people, everyone together, doing the judgment. The candidate can do all of that in one day, right? And you can compact time so much, which becomes incredibly useful and resourceful. That's hiring. And then, if you consider the way to play at scale, it is always about portfolio. The more successful stories, if you understand Sequoia or Y Combinator: take the Sequoia story, you have the founder of Sequoia in the early days working in microchips, understanding microchips were going to be the next big thing. Then he builds a portfolio of the ten companies in microchips he really believed were going to be the best, and from these winners he invests so much more money and generates this portfolio strategy. Maybe that's the same for Stripe, because Stripe did exactly the same, like the DoorDash story. They had all of these customers they were building more tooling for over time, but they knew who was going to be successful and spent more time on it, and then grew into this massive company. Because if you become more useful to some of your customers over time, then you can use this as R and D to go back and build more product, and be more useful over time. So all of this tooling you build internally allows you to move away from this CRM logic and say it is a problem worth being solved in a very different way, so we can really focus on the important things. I usually say that for the CEO it's like doing all the tasks that nobody wants to do, and I'm so happy AI exists to get this out of the equation.

And these are tasks that are difficult to teach someone to do objectively. On paper everyone can do them and it's not the biggest task, but the effort it takes to organise, especially hiring, as you said, is very hard to organise at scale. And even other back-office administration stuff is very hard to organise at scale. From payroll to anything else, having to gather everything, even though there are SaaS and ready-made companies that supposedly do this, is still a burden. The fact that we have this automation internally for these processes and have the time to focus on other things is one thing, but it's also the artifacts that are produced from the process. Now you can collect objective feedback on how that process goes. You can have artifacts, you can have an audit, you have a history of what happened, so it's much easier to build on top of those things and understand them, because the operator is the same, you don't have different operators.

You don't have them in a small company and in smaller teams, and you don't have the internal politics that skew the information or mask it or provide a veil on top so that what happened in the trenches is never exactly known. So now you have the operators and you have the conversation. Now you can take the conversation you had with the model to achieve this and pass it on to the next person, and they have the context. So it's much easier to give work to someone else and have someone replace you when you're away. These things are huge, and they are just the first layer though.

I think it's the idea of the brain, right? It turns out you can build an internal brain in the company, an external brain in the company that does the actual service. It's no longer some sort of Notion, no longer just here are all my processes. It is here's the internal brain that really understands what everyone understands in the company, that everyone can query and have full transparency of what happens in that company, all of the information on the conversations with customers and where everything is moving. So there is information asymmetry plus intelligence on all of the things that used to be chores, and you can allow this to become such an important piece of the business.

I don't remember who said it, but essentially the more senior you become, the more decisions you have to make in a day and the greater the weight of the decisions you make. And one thing that happens is that people, as they grow up in business professionally, get this gut feeling, this seniority. They can make decisions much faster, more educated decisions that carry weight and are propagated across the company. But there is never a decision log of how this happened. Now with AI you have that decision log. So that's where the brain is. You can know why we made this decision, what action we took, and what the outcome was. So if you keep feeding that and keeping that context in, that's how you build the brain. Maybe it's not a company brain, maybe it's Kyriakos's brain or George's brain in context, and they somehow work together. But having that block of decisions, the why and the how and how we came to those decisions, is the biggest value that hasn't come yet but will come in the future, as we operate on these tools for one, two, three years to get that experience out to the tool. Because it's not a matter of the tools growing or the context window growing. It's a matter of how decisions were made, why, and keeping that context as part of the brain.

My addition to that would be that for the majority of MCPs today, from a user experience standpoint, it's not the best, and there are big delays and so on. So it's more of the question of how do we remove technology rather than add technology. By removing processes and by removing technology and by making things easier, the user experience is better, the agent understanding is better and so on.

One good example for this is football leagues. It's the end of the season and you want to see how many points your team needs to take, or what's the worst and best case scenario for something to happen. You can start asking points per play, how many games are left, when does each team play, and gather data and then make a report and see what are the best and worst case scenarios. At every interaction the agent has with the system, that's how it works now with MCPs and skills. So at any particular sub-task that is happening, the intention is never transferred, the original context. So it all happens in parts, which is good, because you can control those parts, you can make those parts secure. But you get a portion of the value that you could get if you transferred the context everywhere. Let's say I'm calling an API to get the points or to get the data for this period in time. The data might not be available. The data might be obscured. Maybe I cannot call that small or that big of a window because it's a performance issue or whatever. If I transfer the context with me and I have a layer in that system that is built for agents, I can actually optimise that query at run time because I know the intent. It's like fetching every row in your database to do a sum, when the database can do that for you. But if you tell the database, or that system, I don't just want the sum, I want to compare the sum to something else, or I want to see if the sum actually changed from last week, and you transfer that context to the system, the system can perform much better for you and be more optimised and perform the task with the context in mind.

So the best way I can simplify this and explain it: the context needs to be carried, and this is part of what teams and people building software have done since the beginning of software. Identify the intent behind the request so that you can build the system in that way. I have a story that I always tell. I don't remember which client I was working with during my web development days. The spec was to make the images in a listing page of a niche shop smaller. So what do you naturally do? You go through some CDN or whatever and you compress the images, you make them a lossless compression so that you have faster response times. The intent was that the stakeholder at the time wanted to make the image a bit fuzzy, a bit blurry.

Why? So that they would force the person to click the zoom-in icon to capture that intent and capture interest on the product. So the spec came out totally different. I didn't know the actual intent. A lot of work was done to make this optimisation and everything. That was back in 2009, I think. With the tools that we have now, the problem could be solved with just the CSS bit, turned on on the spot in five minutes. So you can see that carrying the intent with you is very important.

And this still happens with agents now, with whatever system, and with people as well.

On your second point, can you imagine that in the era of AI you can do your whole balance sheet and the AI does the whole thing? You need approval from an auditor and an accountant to give it to the government. The AI does the whole thing.

What worries me is that the auditor still approves manually and doesn't run AI. This is the biggest thing I worry about.

Yeah. I think we need some sort of a scheme where you get a piece of land and you say I will take every person that leaves the military from Cyprus, and everyone that wants to be there, give them a hundred thousand dollars, or maybe someone that leaves university, all of the youth of the country, and say I'm going to give grants to these people and let them build whatever they want. I'm going to remove all of the processes and remove all of the accountants and lawyers that give out these funds, and enable these people to build. And then ask all of the investors from the valley to come here and tell them, anything you invest, if there is downside, if you lose your money, we're going to give you your money back. So you have the talent there, then second you bring the investors from the outside and you make it risk-free. And then you actually focus, maybe focus on the things that we know we have problems in or that we have a market feeling for. But yeah, it's a tangent that we can go on forever.

Maybe the last one: if you go to Spain and you see the history, you see Alcaraz, he was sixteen years old, he comes up, best player in the world. Maybe in football someone comes up at fifteen or sixteen years old and he's the best in the world. So maybe this has some business sense, maybe this has some coaching sense. But say you wanted to make Cyprus a country where the actual paddle players become really good on a global scale, how would you approach it?

Okay, in countries such as Spain, paddle has been a sport for seventeen, twenty years now. In countries such as Cyprus, it's been a sport for two or three years now. So it's completely different. Even the people that you will play with are completely different. One example is that most courts that are covered in Cyprus have an eleven meter height, which is the minimum requirement. If you meet a player from Spain and you tell them you are at eleven meters height, they will not play. Most of their balls they would hit the roof. They want fifteen meters, thirteen meters. It's a totally different game. So how do you track that? How do you see the progress of a player that goes from the eleven meter court to the fifteen meter court height? You need to see patterns, you need to track what the best players are doing. So you need a way to measure everything and apply the same patterns, observe what you do and see what you need to change. In order for that to happen you need to have a framework of tracking your progress, you need to have the right people to play with and play differently. Because if you play the same people every day you will never get better. So you need that variation, the tracking, the measurement, and all the factors that affect play. Now you start to see gyms that do focused classes and personal training for paddle players, because the movements are different, the muscles you use are different, the explosiveness is different, it's not the same as tennis. The way you run is different. So you need all this knowledge and experience to actually help players progress in that sport. The basics are the same, but the tools to achieve them are now much different. The youth football national team of Cyprus, they don't have trackers on them. The average U16 team abroad, they have trackers. So you can understand the difference you can achieve if you measure it. Cover-based tracking is still just prototypes and it's not there in the important stuff yet. And you have to start young. Most people that start paddle now are over 18. You rarely see kids that are seven playing paddle, but you see a lot of kids that are seven playing tennis. Paddle has a much faster learning curve. Now, as for reaching the top, I wouldn't be able to say. But the younger you start, the more patterns you can see, the more you can apply, the more you can experiment with, and the more you can reach that top level. Starting from youth is probably the most important.

Super. That's an awesome chat, man. Cool.