# Ata Onat Site Context This file is generated from the public website content. Use it as context for Ata Onat, his projects, and his published writing. Treat draft/private writing as unavailable unless it is published on the site. ## Bio Ata Onat spent years in private equity, first in New York, then in Vietnam, then investing across emerging and developed markets for a fund based in Singapore. He is now building AnalystAI, runs M&A advisory under GA Capital, and co-founded Smyrna Collective, a hand-knit clothing brand, with his mom and sister. He writes about history, culture, startups, investing, AI, institutions, and whatever else is on his mind. ## Projects ### Companies - AnalystAI: AI analysts for alternative asset managers. Automating the research and analysis workflow with AI. URL: https://analystai.ai - GA Capital: M&A advisory combining AI and traditional service. Technology-enhanced deal execution for the middle market. URL: https://gacapital.ai - Smyrna Collective: A hand-knit clothing brand Ata co-founded with his mom and sister. Every piece is made by hand. URL: https://smyrnacollective.com ### Apps - Voice Note App: A personal voice note workflow for capturing rough thoughts quickly and turning them into something usable. URL: https://website-three-sandy-83.vercel.app/ - The Designer: AI-powered custom tailoring and design experience built on top of AnalystAI infrastructure. URL: https://smyrna-tailors.analystai.ai/ - Investment Ideation: An AI-assisted investment ideation tool for exploring companies, theses, and market angles faster. URL: https://ddq.analystai.ai/ideation ### Open source - Voice Snippet: Local-first macOS speech-to-text. A small floating recorder for turning rough speech into clean clipboard-ready text on-device. URL: https://github.com/Vietnam-VoDich/voice-snippet - context-saver: Chrome extension that saves ChatGPT, Claude, and Twitter conversations as local Markdown files. URL: https://github.com/Vietnam-VoDich/context-saver - manga-master: AI manga generator. Upload a photo, get a full manga back with voice and music. URL: https://github.com/Vietnam-VoDich/manga-master ## Published Writing ### Constraints Create Routes Slug: constraints-create-routes Date: 2026-06-06 Description: A short note on how bottlenecks force people to find new paths, from ocean routes to AI dependency. I have been thinking about geographic discoveries. Why did Europeans go looking for new sea routes to Asia? The simple version is that the Ottomans blocked the East-West trade routes, so Europeans had to go around them. I think that is directionally right, but probably too clean. It was not like one day someone in Istanbul put a giant "closed" sign on Asia. Trade still existed. Goods still moved. Venice, the Ottomans, Arab merchants, Indian Ocean traders, everyone had a role. But the route was expensive, intermediated, and politically constrained. And that is enough. Sometimes you do not need a wall. You just need a tollbooth, a chokepoint, a dependency, a margin structure that makes everyone else feel stupid for accepting it. Then someone starts asking: can we go around this? That is the part I find interesting. A lot of innovation is not born from abundance. It is born from frustration. The existing route works, but it works on someone else's terms. So you look for another route. Portugal did not wake up one morning and say, "what if we had a fun little ocean adventure?" They were trying to bypass a system. They wanted direct access to gold, spices, India, Asia. They wanted to remove the middleman. The ocean became interesting because the land route was controlled. This pattern shows up everywhere. Oil dependency is bad, so people build alternatives. EVs, batteries, nuclear, solar, efficiency, domestic drilling, strategic reserves. None of these are just cute inventions. They are responses to being vulnerable. If you depend on something critical and someone else controls it, your brain changes. You stop optimizing inside the current system and start looking for an exit. That is why I think the AI situation is interesting. Right now, frontier LLMs are basically a US-China game. Maybe that sounds too harsh, but I think it is true in the strategic sense. The US has the frontier labs, chips, cloud infrastructure, capital markets, talent density, and energy base. China has the scale, state urgency, industrial base, engineering depth, and increasingly very capable models. Everyone else is kind of irrelevant. Mostly America and China. So what does everyone else do? One option is dependency. Just use American APIs. Maybe use Chinese open-source models when convenient. Accept that the most important economic tool of the next decade is imported. But even open-source is not a full escape. A Chinese open-source model still has to run somewhere. That means data centers, chips, electricity, cooling, networking, and operational talent. And where does that usually live? AWS, Azure, Google Cloud, American chip supply chains, or some sovereign cloud that still depends on the same scarce inputs. So the dependency moves around, but it does not disappear. That feels unstable. Countries do not like depending on foreign oil. They will like depending on foreign intelligence even less. And maybe the best-case scenario is that AI becomes a commodity. Maybe models get cheaper. Maybe inference costs collapse. Maybe everyone can access intelligence the way they access electricity or cloud software. That sounds nice. But even then, who is selling the commodity? Mostly the US and China. That is in some ways worse than energy. With oil, gas, coal, uranium, hydro, solar, wind, there are many countries with some form of supply or substitute. You may hate your dependency, but you can diversify it. You can buy from the Gulf, Norway, the US, Australia, Indonesia, Russia, Canada, Brazil, whoever. It is messy, but the map has options. With frontier intelligence, the map may have two real sellers. So even if AI becomes cheap, it may not become geopolitically safe. Maybe non-US and non-China countries can pool resources. Europe, the Gulf, India, Japan, Korea, Singapore, maybe some coalition of capital, energy, talent, and demand. In theory, that sounds reasonable. But even then, I am not sure. These systems cannot simply be copied. Talent is hard to copy. Chips are hard to copy. Energy is hard to copy. Data centers are hard to copy. Research culture is hard to copy. Distribution is hard to copy. The feedback loop between users, products, infrastructure, and model improvement is especially hard to copy. This is why I think the geopolitical implications are huge. I just do not know the solution. So the question is: what is the AI version of the sea route around the Cape? I do not know. Maybe the constraint forces creativity. Or maybe this time a few countries really take off and the gap between the US, China, and the rest of the world becomes meaningfully bigger. You still have places like the Netherlands, Korea, and Japan inside the supply chain, so maybe they stay relevant too. But for countries that are not in the AI supply chain, I am not sure how they stay competitive. Especially if robotics takes off sharply too. That would be brutal for emerging markets. I know this is probably a five- to ten-year horizon. But if things are moving this fast, who knows. That is usually where things get interesting. The Portuguese did not beat the Ottomans by becoming better overland traders. They changed the map. Maybe the next wave of AI innovation will come from the same impulse: not because everyone has equal access to the frontier, but because they do not. Constraint creates rerouting. And sometimes the reroute becomes the new world. ### Earth Has No Exit Slug: no-undiscovered-land Date: 2026-06-01 Description: A short note on frontiers, inherited institutions, and why becoming multiplanetary may matter politically. There is no undiscovered land anymore. That sounds obvious, but I think it matters more than people realize. Every piece of land on Earth is already inside some country. Which means every place is already anchored to a legal system, a regulatory structure, a political history, a culture, a bureaucracy, a set of inherited fights. You do not arrive anywhere as a blank slate. You arrive inside someone else's accumulated past. And a lot of that past was formed before modernity, with assumptions that may not be a great fit for today's realities. The US is a good example of how discovering new land could lead to new ideas. America was not a moral clean slate, of course. It was built on land where people already lived, and it carried plenty of English assumptions with it. But it was an institutional distance machine. Far enough from London to make new assumptions possible. That option barely exists anymore. If you are unhappy with the way a country is run, where do you go? You can move to another country, but that country already has its own system. You can start a company, but the company still sits inside a jurisdiction. You can build a city, but the city still belongs to a state. Even seasteading runs into the same problem eventually: law, sovereignty, recognition, enforcement. There is no outside. And that may be one reason humanity feels politically stuck. Every reform has to happen inside existing systems. Every experiment has to negotiate with inherited constraints. Every new society begins with old paperwork. And that is just hard. This is one reason I find the idea of becoming multiplanetary so interesting. People usually talk about Mars or space colonies in terms of survival: backup planet, asteroid risk, species insurance. That argument makes sense. But I think there is another argument too. Space gives humanity back the frontier. Not just a physical frontier, but an institutional one. A place where new forms of government, property, work, law, and culture could be tried without being immediately absorbed into the accumulated weight of existing states. That does not mean it would be easy. It might be brutal. It might recreate all the same problems. Humans bring their habits with them. There is no guarantee that Mars produces better institutions than Earth. We have the US example, but we also have more exploitative Latin American governance that also came from frontier conditions at some point. So it is not clear that the frontier will always be better than the past. But the possibility matters. Civilization may need frontiers because frontiers create exits. And exits create pressure. If people can leave, existing systems have to compete. If nobody can leave, bad systems can survive much longer than they should. Right now, Earth has no real exit. Maybe becoming multiplanetary is not just about exploration or survival. Maybe it is about giving humanity the ability to start again. ### Apple in China and How Countries Can Learn Slug: capability-is-created Date: 2026-05-27 Description: A short note on Apple in China, supplier development, and why manufacturing capability is built under pressure. One thing that struck me while reading *Apple in China* is that the lesson is not really "Asia had manufacturing capability." That is too simple. In the book, the early factories Apple worked with were often rough. Korea was described as bare-bones and scrappy. A senior Apple engineer thought the systems were crude, with very little high-quality automation. Taiwan was not magically ready either. Quanta's early quality and development capability were subpar, and one engineer described the experience with words like treachery, ineptitude, sloppiness, and negligence. Shenzhen was the same. Fast, enormous, hungry, but the quality was horrendous by Western standards. Buildings were handmade and slapdash. Speed and scale came before quality. So the interesting point is not that Apple discovered fully formed manufacturing excellence in Asia. The interesting point is that capability was created. Apple brought impossible product specs. It brought volume. It embedded engineers directly inside suppliers. Apple veterans said nobody else was sending dozens of engineers into Taiwanese suppliers and pushing what was possible in that way. Chris Novak was sent to Taiwan to help bring up quality and called one factory pretty deplorable. That is the mechanism. Demanding customer + embedded engineers + hard product specs + volume + supplier reinvestment = capability creation. But China was not passive in this. That is also important. The government did not just wait for Apple and Foxconn to show up and magically create capability. It made the whole thing easier. Taiwanese contract manufacturers basically understood that you would be an idiot not to be in China. The labor pool was enormous. Local governments were hungry. Government employees even had quotas to source migrant labor. They helped with land, equipment, permits, infrastructure, and whatever else made scaling faster. So the incentives were huge, but it was not only about incentives. The state was actively trying to remove friction. It made China easier to operate in, easier to expand in, easier to hire in, easier to build in. That matters because supplier development needs an environment around it. Apple could pressure suppliers, but China made it possible for suppliers to absorb that pressure at insane scale. This also connects to something Dan Wang writes about in *Breakneck*, which I totally recommend. China has an engineering mindset. When the goal is clear, that mindset can be extremely powerful. If the objective is to build factories, move labor, expand infrastructure, reduce bottlenecks, increase output, and optimize for a concrete result, China can be terrifyingly good. The system is built to ask: what is the outcome, and what do we need to do to get there? But that same strength can become dangerous when the goal itself is wrong, illogical, or politically distorted. If the target is bad, an engineering state may still optimize very efficiently toward it. That is the uncomfortable part. This is an important distinction. A country or company does not need to begin with world-class capability. It needs the right slope. It needs customers who demand more than it can currently do. It needs people willing to sit inside the factory and transfer knowledge. It needs volume large enough to justify reinvestment. And it needs suppliers hungry enough to absorb the pressure instead of rejecting it. That is how Korea, Taiwan, and China learned. I think about this now with Vietnam. Samsung has been in Vietnam for years, and it clearly matters. There are supplier-development programs. Some capability does transfer. People learn. Managers learn. Local suppliers get pulled into the orbit of a world-class company. But I am not sure it is the same as Apple's historical role in China and Taiwan. The question is how deep the transfer goes. Are Vietnamese suppliers being forced into hard, high-precision, high-volume components? Or is Samsung keeping the highest-value knowledge inside its own system while Vietnam mostly hosts assembly and supporting work? Luxshare raises an even harder question. Luxshare is already a Chinese Apple-trained supplier. If it moves production to Vietnam, does Vietnam learn? Or does Vietnam simply host Chinese-managed assembly? That difference matters enormously. Foreign direct investment is not automatically capability creation. A factory can employ people without teaching the country very much. A supply chain can move geographically without transferring its real knowledge locally. The deeper question is not: are factories coming? The deeper question is: who is learning? That is what Apple seems to have done in China. It did not just buy from suppliers. It helped create them. It forced them up the learning curve, one impossible product after another. That may be the real development lesson. And maybe one has to ask, from the US perspective, whether it was positive in the end that Apple did this. It was certainly positive for Apple shareholders. But a lot of know-how was transferred, and even in the book, Foxconn has Apple train its engineers and then rotates them. So is Korea or China making the same mistake now? I do not know, but it is an interesting question. It is also interesting because people often think FDI is what grows an economy and enables technology transfer. But Apple was not really doing FDI at all. It was doing something stranger and maybe more powerful: supplier development through relentless pressure. ### 1000x Users Slug: the-thousand-times-point Date: 2026-05-26 Description: A short note on AI power users, casual users, and where the real economic impact may come from. I realized something about AI usage. I probably use AI manually two hundred times a day, maybe more. If you include API calls from agents running in the background, the number is probably closer to a thousand. I am sure there are other tokenmaxers out there who do even more. Then I saw some relatives in my hometown using AI. They use it maybe four or five times. A question here, a translation there, maybe help writing something. Useful, but not central to their day. When you spend your time on X/Twitter, you do not realize this gap. But in the real world, for most people, AI is just a replacement for their daily Google searches. This made me think about ChatGPT / Gemini announcements about, say, 900M weekly active users. Well, if one weekly active user has 3 searches, and another has 10,000, that is clearly not a good comparison. If anything, comparison should be based on weekly tokens spent. It is true that tokens are not a direct measurement of productivity either. But the active user concept is going to become less valuable as time goes by, because one active user could have 100 agents. This brings up another important point. Previously, there was a limit to how much leverage a user could have. I might do 100 Google searches, and someone else might do 5. I am 20x him. Now with AI agents, I think I may be 1000x him. Do you see where I am going? AI and agents are adding incredible leverage to heavy users. Who are the heavy users? Coders are one example. Then I thought about my surroundings and who the heaviest users are. They are the people with the most activity in their lives. By activity, I mean people who are trying to get things done. Business owners, managers, perhaps busy mothers. The more things going on in your life, the more you leverage AI, because it helps you start a thread, an initial motion, on a dozen projects you have on your list. Personally, I am currently on a Delta flight, writing this. And I still have to write this myself, because I just cannot like what AI does. Yet in parallel, I have an open-source project for my Google Chrome workflow, a deck for my company, some research on investing and Microsoft, and pretty much 20 other things. So if you are someone who is trying to do a lot, you have infinite opportunities to use AI. And guess what, the start is just a prompt. It is very low effort. I think this has interesting implications for markets, users, and labs. First, people who were already more agentic will use AI more. They simply have more needs for it. Second, they will get more leverage from this technology, which will make them more productive and rich. We have already seen this with Anthropic. Technology does seem to increase the overall pie, while also allocating a bigger portion of it to a few individuals. Third, weekly or monthly active users do not matter as much as weekly token spend or tool calls, because those are better metrics for understanding true usage. A single user may be worth 1000x another. In that sense, you can see that Anthropic was so much smarter than everyone else. They bet on coding before anyone, and they are reaping the benefits. OpenAI realized this, and now with Codex they are close followers too. Google, unfortunately, is still facing the innovator's dilemma. They need cheap models for their billions of users, but those billions of users may soon become less relevant if another set of 10 million users has demand equal to 10 billion people. Another point about heavy users is the type of queries they ask. Most people who use single-shot AI are replacing simple Google queries. In fact, for those, I go to Google too. AI Overviews are super fast and nice. But they are just informational questions, and I do not add much value to Google's underlying models. Whereas with Codex and Claude, I am trying a dozen different ideas. I scream at it, I swear at it, I show where I am unhappy, and I give so much data about what I am trying to achieve. All of this is going through the API. OpenAI / Anthropic are seeing what I, and a million other heavy users, are trying to achieve. They see where the models fail. Then they take that and improve the next iteration. And so it goes, and so the gap widens. I want to be wrong about this, but perhaps this is why Google could not catch OpenAI / Anthropic for a while. Their coding models are simply not as good. Hence the point of this write-up. We are going into a world where 1000x users, 10x engineers, or 100x salespeople will matter more and more. You have to build for those people because they can increase aggregate demand by creating more agents every day. ### Discipline Is Civilization Slug: discipline-is-civilization Date: 2026-05-17 Description: A short note on Rome, barbarians, and why organized societies keep beating disorganized ones. I was reading Gibbon on the decline and fall of Rome, and one thing kept standing out to me. The Romans were not always braver than the people they fought. They were not always physically stronger. In many cases, the barbarians had more raw ferocity, more individual courage, more willingness to throw themselves into battle. But they were less organized. That seems to be the recurring difference. The stronger society is usually not the one with the most intensity. It is the one with discipline. The one that can coordinate. You see this with Rome. Sometimes discipline is embedded in institutions, law, training, hierarchy, roads, logistics, and bureaucracy. And sometimes discipline is imposed by a person. Genghis Khan and Attila the Hun are interesting for this reason. They took groups that could have stayed fragmented and made them coordinate at a much higher level for a period of time. The result was some of the largest empires ever. But the moment that coordination weakens, things can fall back into fragmentation. The founder creates order, but the order does not always survive the founder. That may be the most interesting question: how do you turn the traits of a great founder into institutions? This is why Rome is different. And maybe Japan too, where a warrior class could later become embedded into bureaucracy. The real power is not just discipline for one generation. It is discipline that gets stored in culture and institutions. Gibbon describes this savage fierceness, energy that appears suddenly, burns hot, then turns inward or collapses into disorder. Fierceness without structure. And I think this pattern shows up everywhere. A company with average people who can collaborate beats a company full of brilliant people pulling in different directions. A country with institutions beats a country with talent but no trust. A culture that teaches people to organize will outperform a culture that only teaches people to be impressive individually. Maybe civilization is mostly this: the ability to turn human energy into coordinated action. Raw force matters. Talent matters. Courage matters. But without discipline, they scatter. I don't know of any great civilization, when I read, that looks lazy, sloppy, and undisciplined. Towards the end, they degenerate, yes. But at their prime, they are almost always the ones with more discipline. ### The Slope Matters Slug: the-slope-matters Date: 2026-05-17 Description: A short note on Apple in China, supply chains, and why the rate of learning matters more than the starting point. I was reading *Apple in China*, and there is a pattern in the book that I found interesting. Apple goes through this long supply chain journey. Korea, Japan, Taiwan, China. Each place has problems at the beginning. The quality is not good enough. The process is not tight enough. The teams are not learning fast enough. Something is missing. But the important thing is not the starting point. It is the slope. You can see it in Korea. A company like LG may not have everything figured out immediately, but there is ambition there. There is hunger. They want to learn. You see it again with Foxconn, which is Taiwanese. Same thing. Ambition, speed, willingness to absorb whatever Apple knows and turn it into capability. And then you see it in China. The book describes people working incredibly hard, trying to learn everything, trying to suck in as much knowledge as possible. They are not starting from perfect quality. They are starting from intensity and a willingness to improve. That is the part I find amazing. When you go to a place and it is bad, that does not mean it will stay bad. The question is whether the slope is good. Are people learning? Are they aggressive? Are they dedicated? Are they embarrassed by the gap, or have they accepted it? If the slope is strong enough, the starting point matters much less than people think. You can see this in Korea, Taiwan, and China. They may have started behind, but they were moving. And if a society or company is moving fast enough for long enough, eventually the world wakes up and realizes the gap is gone. ### Twelve Hours Slug: twelve-hours Date: 2026-04-25 Description: A brutal thought from a coffee shop about human time and what we do with it. I am sitting in a coffee shop. At the table next to me, four or five young women have been taking pictures for the past three hours. Posing, adjusting, reviewing, retaking. I don't know when they plan to stop. My brain immediately did the math. Three hours times four people. Twelve hours of human life, right there. I'll be honest: my first reaction was that it's a waste. I know that's a brutal framing. And I sat with it for a while, because I don't think I'm entirely right. Maybe they are genuinely happy. Maybe this is exactly what a good afternoon looks like to them, and who am I to say otherwise. Maybe there is real satisfaction in it - the craft of a good photo, the social ritual of doing it together, the feeling of looking good and knowing it. Those things are real. There's also an evolutionary argument. Looking attractive, projecting an image, attracting attention - these are not trivial activities if you think about them without the judgment. They may be a perfectly rational use of time by a different set of values. Maybe even a winning strategy. And yet. I can't fully shake the feeling. Twelve hours of human time. Four minds. Whatever intelligence, curiosity, or capability sits inside those four people - all of it parked for three hours in service of a photo that will be seen for three seconds on a screen. I don't think this is about these specific women. It's about something larger. We live in a moment where an enormous amount of human energy flows into the production of images of ourselves. Hours and hours every week, across hundreds of millions of people. The cumulative math is staggering. I wonder sometimes what that energy could produce if it ran somewhere else. Not in a moralistic way - I'm not interested in telling anyone how to spend their afternoon. More in a curious way. What problems go unsolved, what ideas don't get thought, what things don't get built, because this is where the hours went instead. Maybe nothing. Maybe the hours would just go somewhere else equally unproductive. Probably. But still. Twelve hours. Gone, just like that. ### The AI-Native PE Firm Slug: the-ai-pe-firm Date: 2026-04-07 Description: Why I tried to build an AI analyst, why it failed, and where I think the real opportunity actually is. For the past two years I've been building in the AI for finance space. My first thesis was an AI analyst, and on that thesis my timing was off in both directions. First I was early. Then I was late. Along the way the product evolved into something I'm actually proud of, and the experience taught me where the real opportunity in this space is going to be. The early version The pitch was simple. A real analyst at a PE firm or investment bank costs $100K to $120K all-in. What if we could give you something that did the work for \$20K? The problem was that the agent wasn't actually an analyst. It looked like one in a demo. It collapsed in real work. I'd watch the supposed analyst write up a potential buyers list for a deal and the analysis would be wrong in ways that mattered. Not factually wrong. Wrong in nuance. It would miss the thing an actual analyst would catch in five minutes because they understand how the industry works, who the real strategics are, what a fair multiple looks like in this corner of the market. The model knew some things really well and other things not at all. And it had no idea which was which. That's the brutal part. Confident wrongness is worse than knowing nothing. A junior analyst who doesn't know something asks. An agent that doesn't know something writes a paragraph. You can also see this in the pricing. Most AI analyst tools are priced at $200 a month, not $20K a year. If they were really doing the work of an analyst, the market would clear at the higher price. It doesn't. The market is telling you what these things actually are. The agents are much better now. They have more nuance. But by the time they got good enough, the second problem appeared. The late version The companies building in this space now are not bootstrapped startups. Anthropic and OpenAI are hiring ex-investment bankers. Harvey has raised hundreds of millions. The frontier labs are doing direct deals with the largest financial institutions in the world. This is not a space where a small team is going to win on insight. It's a space won by capital and distribution, and both have already arrived. That specific thesis (sell an analyst replacement) is dead, at least for me. But the work didn't stop there. Over time the platform evolved into something broader: it helps on the deal side, but also after the deal closes. Automated scrapers that track what a portfolio company's competitors are doing. A buyer finder for when a portfolio company needs a strategic exit. Tools for risk and compliance, market tracking, operational monitoring. The interesting thing is that none of this looks anything like the original "AI analyst" pitch. It looks like the connective tissue of a modern investment firm. And that's what pointed me toward where I think the real opportunity actually is. Where the real opportunity is Everyone in AI for finance is building on the deal side. Sourcing, due diligence, document parsing. Why is everyone there? Because that's where the data is clean. You connect to a data room and you get a folder of PDFs. It's the most structured part of the business, which makes it the most tractable. The harder, more valuable opportunity is the opposite end of the firm: portfolio operations. When a PE firm buys a company, the real value creation doesn't happen in the deal. It happens in the next three to five years inside that portfolio company. Margin expansion, working capital, sales productivity, operational efficiency. Almost none of this work is being touched by AI right now. The reason it's not being touched is the reason it's valuable. Portfolio companies are messy. Every business has a different topology. You can say "logistics companies are all similar" until you walk into one and discover their inventory taxonomy was invented by someone who left in 2008. There's no clean data room. There's a CFO who exports things to Excel and an ops manager who keeps everything in his head. That mess is exactly why the opportunity exists. You can't build one product and sell it to a thousand companies. But a PE firm doesn't need to. It needs to deliver outsized returns across 5-7 portfolio companies, which is a completely different shape of problem. In-house, not consulting The instinct will be to outsource this to consultants. Bring in a boutique, run an assessment, deliver a deck, charge a fee. This will not work. It won't work because consultants give a view and then leave. The portfolio CEO nods, says "great, very interesting," runs one fake project to satisfy the fund, and goes back to running the business the way they always did. Real implementation requires skin in the game. PE firms already understand this for other functions. Every serious mid-market PE firm has an in-house operating partner team that goes into portfolio companies for six to nine months after a deal closes. What I'm describing is the same model with AI-native operating partners instead of supply-chain or finance ones. You parachute in a small team that has actually shipped real systems. They identify the three or four places where AI can create real ROI in this specific business. They build it, ship it, measure it, and move to the next portfolio company. The moat isn't the technology. It's that the implementation is hard, the talent is scarce, and the value is captured by the fund instead of leaking out to a consultant. Why almost no one will do this I think this will get built. The interesting question is who builds it. Most PE firms are run by excellent dealmakers. Dealmakers do deals. They are not spending their weekends building with the latest models, and they shouldn't be. That's not their job. But it means very few people sitting at the top of these firms will recognize the right AI talent when they see it, and the right talent won't want to work for someone who doesn't get it. I think in five years a small handful of PE firms will operate this way and outperform the rest by a meaningful margin. Many more will try, hire the wrong people, run a few pilots, and quietly go back to their old playbook. The execution gap will be enormous. Most of the value will go to the few who actually do it. ### Base Models and Upbringing Slug: base-models-and-upbringing Date: 2026-04-06 Description: On how reinforcement learning shapes AI personality the same way education shapes people. In AI model training, there's always a base model, and then there's the reinforcement learning on top of it. If you've paid attention to how these AI models have evolved, this distinction matters more than most people realize. Think about the early ChatGPT models. GPT-4o, the original Gemini releases. Those models were incredibly agreeable. They supported your view, validated your ideas, told you what you wanted to hear. People loved it. Some people got genuinely addicted to it. That's what the industry calls sycophancy. Then OpenAI tried to fix it. The series leading up to GPT-5.4 swung hard the other way. The model became almost robotic. Super technical language, no warmth, no conversational flow. You couldn't even tell you were talking to something designed for humans. It felt like talking to a textbook. The more I think about this, the more I see a parallel to how people develop. Imagine a kid with the same evolutionary hardware as everyone else. Then you send that kid to a deeply religious school for a few years. You see them change completely. I have seen this when I was growing up with some friends. Education and environment could really reprogram you. The upbringing, the highly specific environment overrides the baseline. That's essentially what's happening with these models. Pre-training a base model is doing human evolution on fast-forward. Each training run is like sending a human through 100 million years of evolutionary pressure to build the deep substrate, the raw capability to survive and understand. But then comes the reinforcement learning, the fine-tuning. This is the upbringing. It's the education. In humans, I feel like training and education end up being more important in shaping the final person than the evolution itself, though to be fair, I haven't seen what we evolve into in another 10 million years. I'm not saying the analogy is perfect. These systems are far more complicated than that. But the pattern is striking and maybe, similar to how we have different nationalities and cultures, perhaps the same will happen with AI models. Some will be more agreeable, some less, some are more conscientious (why Claude always says let's stop here and continue later, for instance?) and some are just like autistic debuggers. ### The Open-Source Moment Slug: the-open-source-moment Date: 2026-04-06 Description: On running Gemma 4 locally and realizing the threat to closed AI models might be real. I've been running Google's Gemma 4 model on my laptop since Friday. It came out April 3rd and I installed it the same day. I'm quite impressed. For basic knowledge tasks, summarization, answering questions, it works. You ask something and you get an intelligent answer. Tool calling was a bit rough at first, but it works too. Tool calling, for context, is how an AI model interacts with external services. Think of it like someone who can use Excel. Excel is the tool. The model decides when to open it and what to do with it. Here's what made me think. This model is free. It runs on my machine. My files never leave my laptop. There is no API bill, no subscription, no data going to someone else's server. And for a lot of everyday tasks, it's genuinely good enough. Now, I still think you need the best closed models for coding. That's where the gap is real. When you're building software, the difference between a good model and a great model is enormous. But for everything else? Research, summarization, drafting, analysis? The local model put me in a real dilemma. People talk about open-source models being six to nine months behind the frontier. That sounds like a lot, but think about what that means in practice. The model I'm using for free today is roughly where the best paid models were last fall. And last fall's models were already very good. The hardware keeps improving too, though on longer cycles. Two to three years for meaningful chip generations. I wouldn't be surprised if in two to three years I can run something equivalent to Opus 4.6 on my Mac. For free. And Opus 4.6 is already an incredible model. Which raises an interesting question for the closed-model companies. If open source keeps closing the gap, the frontier models have to keep getting dramatically better to justify the cost. But what if they don't? What if we're approaching a point where the improvements become incremental rather than transformational? Then the value proposition of paying for a closed model gets harder to defend for most use cases. I think this was the first time I really felt it. Everybody talks about open source as a threat to the big AI companies. I've read the arguments, I've seen the benchmarks. But this was the first time I actually used an open-source model seriously, experimented with it for real work, and thought: wow, this is actually a threat. There's a difference between understanding something intellectually and feeling it. Friday was when I felt it. ### The Restaurant Reset Slug: the-restaurant-reset Date: 2026-04-05 Description: On how a good meal can make you forget everything, and what that means about hospitality. Sometimes you're in your worst moment. Your business is about to go bankrupt. Nothing is working. You haven't slept properly in weeks. The weight of everything is just sitting on you. And then you walk into a restaurant. The food is incredible. The dessert is incredible. The whole experience just lands. And for a moment you sit there thinking, "Wow, life is actually good." You forget everything. The stress, the fear, the spreadsheet that doesn't add up. Gone. Just for a bit, but gone. I find that pretty remarkable. Someone you've never met, who knows nothing about your situation, just gave you one of the best moments of your week. Maybe your month. Not by solving your problems, not by giving you advice, but by making really good food and putting it in front of you with care. That's what hospitality is, when you think about it. It's not just service. It's the ability to reset someone's day. To take a person who walked in carrying the world and give them an hour where none of that matters. And the thing is, this isn't unique to restaurants. Any product, any experience that's made with real care has this power. When you build something good, genuinely good, you don't always know who's on the other side. You don't know what kind of day they're having. But you might be the thing that turns it around. That's a pretty cool thing to be able to offer other people. ### Capital Is Not Conviction Slug: capital-is-not-conviction Date: 2026-04-03 Description: On FBA aggregators, bandwagon investing, and why billions of dollars flowing somewhere doesn't mean it's the right direction. A few years ago, the FBA aggregator model was the hottest thing in e-commerce. The idea was simple and, on paper, very private-equity-friendly: buy a bunch of small Amazon sellers, aggregate them under one roof, cut costs by centralizing SG\&A, and scale. I come from an investing background, so I looked at it seriously. I thought I had the right skills: evaluating businesses, buying them, optimizing operations. I spent a lot of time on it. But two things kept nagging me. First, the founders of these small Amazon businesses were doing a lot of jobs that you can't easily replicate with employees and managers. They were obsessive about their products, their listings, their customer relationships. They wore every hat because they cared, not because they couldn't afford to hire. When you replace that founder with a salaried operator inside a portfolio company, something breaks. Second, the real skill in e-commerce isn't aggregation. It's developing new products. That requires ingenuity, a feel for the market, a kind of scrappy R\&D that doesn't come naturally to the traditional buyout playbook. You can optimize costs all day, but if you can't create the next product, you're just managing decline. I was bearish. I wrote it up, left a note (I always do that when I pass on something) and said: I don't think this model works. Then I watched companies in the space raise \$3.5 billion. From very smart investors. Silver Lake, Goldman Sachs, and others. These are not dumb people with dumb money. And then they went bankrupt. I'm not picking on any one company here. This pattern repeats across industries. Vector databases were the same story two or three years ago, everybody piled in because it felt adjacent to AI and the momentum was irresistible. Now most of those bets look questionable. Here's what I think is happening. The top-tier investors (the very best ones) sometimes do have a genuinely unique thesis. But most don't. What most have is access, brand, and the ability to follow. There's a massive bandwagon effect in venture capital and growth equity. The top guy goes in, and then everyone else scrambles for exposure to the same theme. Not because they independently arrived at the same conclusion, but because they don't want to miss it. The result is that billions of dollars can flow toward something that is fundamentally wrong. Capital is not conviction. Capital is often just momentum dressed up as a thesis. For founders, I think this is actually an important point. Don't jump on something just because money is flowing there. That signal is weaker than you think. I'm glad I didn't jump on several opportunities like this that I came very close to. Not because I'm smarter than the people who did. I just couldn't convince myself the thing actually worked, and it turned out that was enough. I felt AI was working, jumped in it, and so far we are not a big success, but I still believe in the technology and how it will change the world. ### The Coconut Kingdom and the Branding Problem Slug: coconut-kingdom-and-the-branding-problem Date: 2026-04-03 Description: On Ben Tre, emerging markets, and why great products don't always become great brands. When I was doing industrial real estate in Vietnam, I visited a region called Ben Tre. I had lived in Vietnam for four years at that point and somehow had no idea what I was about to walk into. Everything in Ben Tre is coconut. Coconut candy, coconut crackers, coconut bread, coconut oil, coconut everything. The entire local economy orbits around this one fruit. It was honestly kind of wild, and fascinating. It got me thinking about something I've noticed across emerging markets: they often have incredible products but terrible branding. The raw material is there, the quality is there, the tradition is there. What's missing is the story. Turkey is the same way. Turkey produces some of the best olive oil in the world. Turkish pistachios are extraordinary. But if you asked most people in Europe or America, they'd associate olive oil with Italy and pistachios with... also Italy, or maybe Iran. Meanwhile, Italy took Turkish-level pistachios and built an entire pistachio cream industry that everyone knows and loves. The product isn't necessarily better. The branding is. Why does this keep happening? Part of it is historical. Many of these countries weren't open to global markets for long stretches. Part of it is that branding requires a kind of cultural confidence and marketing infrastructure that takes time to build. And part of it is just that nobody thought to do it. If I had the chance, here's what I'd do in Ben Tre: build a massive coconut-shaped building right in the center of the region. Make it impossible to miss. Turn it into a destination: a museum, a tasting center, a place where tourists and buyers from around the world come to experience what Ben Tre coconut actually is. Heck, do a coconut oil shower. Who cares! In this Instagram world, that would really fly. That's not a factory. That's a brand. And honestly, I just want more Ben Tre coconut in my life. Even a Ben Tre coconut ice cream is just amazing. Emerging markets don't have a product problem. They have a storytelling problem. And the ones that figure out how to tell their story first are going to win. ### Crisis Is the Catalyst Slug: crisis-is-the-catalyst Date: 2026-04-03 Description: On Hamilton, Napoleon, and why the greatest leaders always emerge from the worst moments. The more I read about great leaders, the more I notice the same pattern: they don't rise in calm times. They rise in chaos. Hamilton was an orphan from a Caribbean island with nothing. He arrives in New York right as the colonies are about to break from Britain. Without that revolution, he's a smart kid with no connections who maybe becomes a lawyer. Instead, he becomes Washington's right hand, designs the financial system of a new country, and changes the course of history. The crisis made the man, or at least, it gave the man the opening. Napoleon is the same story. He's a minor Corsican noble, an artillery officer with a thick accent that Parisian elites laughed at. Then the French Revolution tears the entire old order apart. Every established general is either dead, exiled, or discredited. Suddenly there's massive, unprecedented room for a young officer who's competent and ambitious. By twenty-six he's commanding the Army of Italy. Without the revolution, he retires as a colonel if he's lucky. Ho Chi Minh spent decades wandering (Paris, Moscow, Canton), organizing, writing, waiting. He was a revolutionary without a revolution. Then World War II happens, Japan occupies Indochina, France collapses, and suddenly the entire colonial order is shattered. He walks into that vacuum and declares independence. The crisis didn't create his ideas, but it created the conditions where his ideas could actually win. Churchill is maybe the clearest case. He was politically finished by the late 1930s. A washed-up backbencher who'd been wrong about India, wrong about the abdication, too aggressive, too old-fashioned. Then the war comes, and everything that made him difficult in peacetime (the stubbornness, the rhetoric, the refusal to compromise) becomes exactly what Britain needs. The crisis didn't just open the door for him. It made his specific qualities, which were liabilities in normal times, into the most valuable assets in the country. Alexander the Great inherited a throne because his father was assassinated. If Philip II had lived another twenty years, would Alexander have conquered Persia at twenty-five? Would he have conquered it at all? Philip was already the most powerful man in Greece. Alexander might have spent his prime waiting in the wings, growing restless, maybe even growing complacent. The pattern is always the same: a crisis destroys the existing order, and in the wreckage, young and hungry people rise faster than they ever could have in stable times. The old guard is gone or weakened. The rules are suspended. The people who thrive are the ones who are agile enough to move when everything is moving. You can see it in the American founding. These were mostly young men: Hamilton was in his twenties, many of the others in their thirties. In a stable British colonial system, they'd have been merchants, lawyers, minor politicians. The revolution compressed decades of career into years. It took people who would have been notable and made them historic. I think about this a lot, actually. Not just as history, but as a lens for right now. Whenever there's a major disruption (a war, a financial crisis, a technological shift), the instinct is to see it as a threat. And it is. But it's also the moment when the next generation gets its opening. The old playbook stops working, and the people who can write a new one fastest are almost always younger, hungrier, and less attached to how things used to be. Perhaps AI is that moment for Gen Z, who knows. Crisis doesn't create greatness. But it creates the conditions for greatness to emerge. So if you are living in a country or in an environment where things are going bad, perhaps having this perspective would give you some energy to try things. ### If You're an Investor, You Should Play with AI Slug: investors-should-play-with-ai Date: 2026-04-03 Description: Why actually using AI tools - not just reading about them - gives you a real investing edge. If you invest in anything (public markets, private markets, whatever), you need to be playing with AI. Not reading about it. Not watching demos. Actually using it, building with it, getting your hands dirty. Here's why. Your mental model is already stale The world is changing so fast right now that your understanding of what's possible will not update unless you're using these tools. And I don't just mean ChatGPT or Claude in a chat window. I mean coding agents: Claude Code, Codex, the tools that actually let you build things and automate workflows. They're more flexible, more revealing, and they'll change how you think about what software can do. If you're still forming your AI thesis from analyst reports, you're behind. You'll see gaps the market doesn't When you use the APIs directly, you develop intuitions that are almost impossible to get secondhand. Here's a real example. About a year and a half ago, I was using Google's API heavily (their 1.5 Flash model at the time). It wasn't great at coding, but it was surprisingly good at summarizing, scraping data, and creating structured outputs. It was efficient and it was cheap. Meanwhile, the market was absolutely hammering Google's stock. The consensus was that they were hopelessly behind in the AI race. But I was sitting there thinking: these guys are not that far away on the model game. They clearly had an execution problem, not a model problem. The underlying capability was decent and improving fast. I also used Serper a lot for data scraping. And if you look under the hood, Serper is going to Google for its results anyway, another small signal that Google's infrastructure was more embedded than people realized. None of this was a sophisticated thesis. I wasn't deeply technical on model architectures. But these small data points from actually using the products gave me a different view than the market had. I took a sizable Google position and it worked out really well. Tokens are a lens on the economy When the first thinking models came out (the ones that reason step by step), I noticed something obvious if you were paying attention: they use a lot more tokens. Way more. And if models are consuming dramatically more tokens per session, that means more compute, which means more demand for memory. So I bought memory stocks. I actually begged a friend in Korea to buy SK Hynix for me because I couldn't buy it from the U.S. That worked out too. The edge is in the using I had a chance to do a few more similar trades in the past 18 months. When you use these tools daily, you develop a feel for what's real and what's hype, which products are genuinely useful and which are demos dressed up as products. Right now, for instance, Microsoft is getting hammered. I use Microsoft's APIs and products regularly. And this time, I don't have the same contrarian conviction I had with Google, however. The point is simple: being close to the technology and actually using it will give you insights you cannot get from the outside. If you're doing any kind of investing right now, public or private, this is probably the single biggest edge available to you. And it's free. You just have to start building. ### The Organizing Instinct Slug: the-organizing-instinct Date: 2026-04-03 Description: Why Americans are so good at forming clubs, and what the rest of the world might be missing. When I was in the U.S., first at Michigan, then in New York, I kept running into something I'd never really encountered in Turkey: the sheer density of organized groups. Fraternities, clubs, societies, think tanks, heritage foundations, policy organizations, alumni networks. In New York you have places like the Union Club. On campuses you have Greek life, student government, a club for literally everything. And then beyond the university, there's this entire ecosystem of organizations: conservative ones, progressive ones, industry-specific ones, neighborhood ones. People just... form groups. Constantly. At first I thought this was an elite thing. Country clubs, old-money institutions, invitation-only circles. And sure, that exists. But what impressed me was how far down it goes. This isn't just for the upper class. Average people in America join things. They organize. A high school kid runs a club. A neighborhood has an association. Veterans have their lodge. Parents have their PTA. It's everywhere. I never saw this in Turkey. Maybe it exists in some form (there are professional chambers, some social clubs), but nothing close to the same scale or culture. And I suspect most of the East is the same. People gather informally, of course. Family networks are strong. But this specific thing, the instinct to formalize a group, give it a name, elect officers, set a mission. That feels distinctly American. Maybe parts of Europe have it too, but not at the same intensity. What struck me is what this actually teaches people. If you're a teenager and you have to run a club (recruit members, organize events, manage a budget, deal with disagreements), you're learning to organize. You're learning leadership not from a textbook but from doing it badly at seventeen and then doing it slightly less badly at nineteen. Multiply that across an entire country and you get a population that knows, almost instinctively, how to bring people together around a shared purpose. That's a real skill. And I wonder how much of America's effectiveness in business, in politics, in civil society comes from this one cultural habit: the habit of organizing. Most countries don't train this. They teach people to study, to work, to compete individually. But the ability to say "let's form a group and go do something," that might be one of the most underrated advantages a society can have. ### The Token Dependency Slug: the-token-dependency Date: 2026-04-03 Description: We used to worry about oil dependency. Token dependency will be worse. The world runs on oil. We know this. Wars have been fought over it, economies have been built around it, entire regions derive their geopolitical power from it. The Gulf states matter because of energy. Russia's leverage is its pipelines. This is obvious. What's less obvious, and what I think people are massively underestimating, is how dependent the world is about to become on tokens. I'll start with myself. I'm maybe twenty times more productive with AI than without it. I wasn't a great coder (I was self-taught), and now I have six or seven AI agents running at any given time, even when I'm on my phone. Even this essay: the first draft started as a voice recording, which an AI transcribed, and then another agent converted into an outline. My entire workflow, from research to writing to building products, runs through AI models. And I'm just one person. Now imagine this at scale. We're moving toward a world of agentic AI: systems that run 24/7, autonomously handling tasks, making API calls, processing data, generating outputs. Companies will depend on these agents the way they currently depend on electricity. Except electricity comes from diverse sources and local grids. AI models come from a handful of companies in two countries. Think about what that means. If you're calling these models from an API, your entire operation runs through someone else's infrastructure. Your productivity, your competitiveness, your economic output: all of it flows through a pipe controlled by OpenAI, Anthropic, Google, or their Chinese equivalents. Now think about sanctions. The U.S. can already sanction countries out of the global financial system. That's powerful but messy: it requires coordinating with banks, monitoring transactions, dealing with workarounds. Sanctioning AI access would be cleaner and more devastating. You just turn off the API. One switch and a country's most productive tools go dark. No workaround, no alternative routing, no smuggling. This is more effective than any oil embargo. You can stockpile oil. You can't stockpile tokens. This is why local models matter so much more than people realize. Every country outside of the U.S. and China has a massive incentive to invest in open-source, locally-hosted AI models. Not because they're ideologically committed to open source, but because it's a matter of sovereignty. During the Cold War, there were two blocs (the communist world and the West) and then there were the non-aligned countries. The non-aligned movement existed because smaller nations recognized that full dependence on either superpower was dangerous. I think we're heading toward something similar with AI. Right now, China is giving away powerful open-source models, which is amazing. But you can't assume they'll keep doing it. Generosity in geopolitics is always strategic, and strategies change. If you're a mid-sized country building your economy on Chinese open-source models, you're one policy shift away from losing access. What the world probably needs (and I'm genuinely curious to see if this happens) is something like a supranational entity that funds and maintains open-source AI models at near-frontier capability. A collective effort by the countries that are neither the U.S. nor China to ensure they always have access to competent local models. Not necessarily the absolute best models, but close enough that losing API access to the frontier labs doesn't cripple your economy. This might sound dramatic. But if you told someone in 1970 that a group of oil-producing countries could hold the entire global economy hostage by restricting supply, that would have sounded dramatic too. And then 1973 happened. If we thought the world was already dominated by the U.S. and China, I can almost guarantee that in fifteen to twenty years it will be far more dominated, because these will be the countries with full control over the smartest systems ever built. Every other country will either have hedged with local models, or they'll be completely dependent. And dependency, as history keeps reminding us, is never free. ### Think of It Like a Geometry Problem Slug: think-of-it-like-a-geometry-problem Date: 2026-04-03 Description: A habit from Turkey's university entrance exam that still helps me get unstuck. When I was preparing for the university entrance exam in Turkey, I fell in love with analytical geometry. If you know the exam, you know geometry is different from the other subjects. You cannot get through it on memorization alone. You have to be a little creative. You'd get a triangle with a circle inscribed in it, or some configuration that looks completely stuck, and the way to solve it is not to apply a formula. It's to draw one line. Find an angle. Rotate something in your head. And suddenly, through that one move, you can calculate everything. I always loved that. There's something satisfying about the way geometry works. Once you see the right line, the angles fall into place, the side lengths follow, and the whole thing just opens up. I still think about this whenever I'm stuck on something. I have a few notes I come back to, and one of them is just: think of it like a geometry problem. Because that's what being stuck usually feels like. You're staring at something and you see nothing. No way in. No obvious move. And the instinct is to stare harder, to try to force it. But in geometry, forcing it doesn't work. You have to try a different line. Draw something that wasn't there before. Approach from a different angle, literally. The moment you do, the problem doesn't just get easier. It becomes obvious. You wonder why you couldn't see it before. I think this applies to almost anything. A business problem that looks intractable. A conversation you don't know how to have. A decision where every option seems wrong. Most of the time you're not missing information. You're missing a line. One shift in perspective and the whole thing unlocks. Draw the line. Find the angle. The rest follows.