PJFP.com

Pursuit of Joy, Fulfillment, and Purpose

Tag: Sequoia Capital

  • Paul Graham in Stockholm on Why Founders Should Go to Silicon Valley and How Sweden Can Become the Silicon Valley of Europe

    Paul Graham, the Y Combinator co-founder whose essays have shaped how a generation of founders thinks about startups, took the stage in Stockholm to answer two questions at once. Should you, as an ambitious founder, go to Silicon Valley? And what should Sweden do to thrive as a startup hub? His surprising thesis is that both questions have the same answer. Watch the full talk on YouTube.

    TLDW

    Graham argues that talent in any high-intensity field concentrates in one geographic center, the way painting clustered in 1870s Paris, math in Gutting around 1900, and movies in 1950s Hollywood. For startups today, that center is Silicon Valley. Founders should go, at least for a while, because the talent pool is both bigger and better, because serendipitous meetings outperform planned ones, because investors decide faster, because moving abroad paradoxically earns more respect from investors at home, and because measuring yourself against known greats like Brian Chesky, Sam Altman, or Max Levchin clears away the fog at the summit and shows you the work required to get there. The most subtle benefit is cultural. Silicon Valley has a 60 year old pay it forward custom in which people help strangers for no reason, a habit Graham traces to a place where nobodies become billionaires faster than anywhere else. The pivot to Sweden is that the best way to help Stockholm become a startup hub is for Swedish founders to go to Silicon Valley, ideally through YC, and then come back, importing money, skills, and Valley culture. Yes, returning founders are only half as likely to become unicorns as those who stay, but selection bias and the valuation gap explain most of that, and half a unicorn is still extraordinary. The job of Silicon Valley of Europe is unclaimed. Mountain View was a backwater in 1955 too. Critical mass is invisible until it is reached.

    Key Takeaways

    • Whenever humans work intensely on something, one place in the world becomes its center. Painting in 1870 was Paris. Math in 1900 was Gutting. Movies in 1950 was Hollywood. Startups today is Silicon Valley.
    • Every ambitious person working in those eras faced the same decision founders face now. The right answer is the same one it has always been. Yes, go. You can come back, but you should at least go.
    • National borders do not change the basic logic of moving from a village to a capital city. The reasoning that says move to where your peers are does not even know the dotted line on the map is there.
    • At the great center, the talent pool expands in two dimensions at once. The people are better and there are more of them, and they cluster, producing an intoxicating concentration of ability.
    • Serendipitous meetings are mysteriously, enormously valuable. Biographies of people who do great things are full of chance encounters that change everything.
    • Graham offers three candidate explanations for why unplanned meetings beat planned ones. There are simply more of them, so outliers are statistically unplanned. Planned meetings may be too conservative because they require a stated reason in advance. Unplanned conversations let you bail in the first few sentences, so the ones that continue are pre filtered for fit.
    • For ambitious people there is nothing better than serendipitous meetings with other people working on the same hard thing. Big centers produce more of them.
    • Things move faster in big centers because better people are more confident and more decisive, and because peers compete with and egg each other on. Ideas get acted on rather than half held.
    • Investors in Silicon Valley decide dramatically faster than European investors. They are more confident and they face stiff competition, so they cannot sit on a good opportunity without losing it.
    • This produces a counterintuitive rule. The more right an investor is about a deal, the less time they can wait, because everyone else who meets the same founder is going to invest too.
    • Yuri Sagalov is the canonical example. He invested in Max Levchin instantly because he knew anyone else who met Max would invest. Speed is the rational response to a crowded, high quality market.
    • Valley investors grumble that valuations are too high and decisions too rushed, yet they outperform European investors empirically. The complaining is just noise.
    • Moving abroad earns you more respect from investors back home. Jesus said no one is a prophet in their own country, and local investors implicitly assume local startups are second rate everywhere, not just in Sweden.
    • Leaving inverts that rule and lifts you in local investors estimation. Sometimes the mere announcement that you got into Y Combinator is enough. Investors who ignored you for months suddenly trip over themselves to write checks.
    • The Dropbox story illustrates this perfectly. A big Boston VC firm spent a year offering Drew Houston encouragement and advice but no money. The moment Sequoia got interested in Silicon Valley, that same firm faxed Drew a term sheet with a blank valuation. Drew went with Sequoia anyway and in 2018 Dropbox became the first YC company to go public.
    • The biggest advantage of moving to a great center is not what it does for you but what it does to you. A big fish in a small pond cannot tell how big it actually is.
    • In a big pond you can measure yourself against known giants. Surprisingly often the news is good. You see Brian Chesky or Sam Altman or Max Levchin and realize they are not a different species. You could do what they did if you worked that hard.
    • The key word is hard. Seeing a giant up close also calibrates the cost. It is not just I could be like that. It is I could be like that if I worked as hard as that.
    • Graham offers a Mount Olympus metaphor. Moving to the mountain clears away the fog at the top. The summit is right there, quite high but no longer impossibly high. Ambitious people need a high but definite threshold.
    • The most surprising thing about Silicon Valley to outsiders is that people help you for no reason. A founder who recently moved from England said every conversation seems to end with what can I do to help you.
    • This is not politeness. English people are far more polite than Americans on average. The helpfulness is a different cultural artifact specific to the Valley.
    • Graham traces the origin to economics. Silicon Valley is the place where nobodies become billionaires faster than anywhere else, so being nice to nobodies has historically paid off. If the helping behavior was ever calculated, the calculation is gone now. The custom is 60 years old and has become reflex.
    • Ron Conway is the purest expression of the pattern. All he does is help people. He does not track whether they are portfolio companies. He does not remember most of the favors. That untracked, indiscriminate helpfulness lets him operate at a much larger scale.
    • When many people behave this way at once, the conservation law for favors breaks down. There are just more favors. The pie grows.
    • Moving to the Valley changes you. One of the strangest effects is that it makes you more helpful to other people.
    • The answer to how Sweden should thrive as a startup hub is buried inside the answer to whether founders should go. Go to Silicon Valley for a bit and then come back.
    • That move helps Sweden in three concrete ways. The average quality of Swedish startups goes up. Returning founders bring Silicon Valley money back with them. And they import Silicon Valley culture, which has spent decades evolving to be optimal for startups.
    • Silicon Valley culture is more compatible with Swedish culture than people realize. Sweden lacks the tall poppies problem (which it should drop anyway) and shares the high trust trait that makes the Valley work.
    • Historical precedent backs this. In the 1800s Sweden literally gave mathematicians fellowships conditional on leaving the country to study math abroad. Boycotting Gutting in the name of building Swedish math would have been absurd.
    • YC is the optimal way to do the go for a bit and come back move. It is a deliberately engineered super valley within the Valley, concentrating density of founders, helpfulness, and investor speed into four to six months.
    • If the Swedish government designed a program to give Swedish founders concentrated Silicon Valley exposure, they could not do better than YC, and it costs them nothing because Silicon Valley investors fund it. They do not even have to license it. They just call the API.
    • YC data shows founders who go home are only about half as likely to become unicorns as those who stay. Three reasons not to be discouraged. First, selection bias. The most confident and determined founders are the ones willing to relocate, so the data is measuring those traits as much as Valley effects.
    • Second, the metric is valuation, not company performance. Bay Area startups simply raise at higher multiples for the same business.
    • Third, even half as well is still very good. If you would have been a Valley billionaire and end up with 500 million instead, the practical difference is zero. In Swedish kroner you are still a billionaire.
    • Money is not everything anyway. Once you have kids, where they grow up becomes the dominant question. That is an argument for returning home that has nothing to do with startups.
    • The most exciting upside is that Stockholm could become the Silicon Valley of Europe. The job is unclaimed. Nobody has a confident answer to where the European tech center is.
    • Geographic size is not the constraint people think it is. Mountain View was a backwater in 1955 when Shockley Semiconductor was founded there, and it stayed the geographic center of Silicon Valley until 2012 when activity shifted to San Francisco.
    • The two ingredients required are a place founders want to live and a critical mass of them. Stockholm clearly clears the first bar. The second is impossible to measure until you hit it, at which point it tips quickly.
    • Stockholm may be closer than it looks. Critical mass is the kind of threshold that is invisible until it has already been passed.

    Detailed Summary

    Why Centers Exist and Why You Have to Go There

    Graham opens with a historical pattern. Whenever a field gets pursued intensely, one place becomes its center. Painting in 1870 was Paris. Math in 1900 was Gutting. Movies in 1950 was Hollywood. For startups now it is Silicon Valley. The question every ambitious person in those eras asked, should I go, has had the same correct answer for thousands of years. Yes. You can come back, but at minimum you should go. The logic does not change at national borders. If a villager interested in startups would obviously move to their country’s capital, the same reasoning applies when the capital sits across a dotted line on a map.

    What you get at the center is a talent pool that expands in two dimensions at once. The people are better, and there are more of them, and they cluster, producing a density of ability that Graham describes as intoxicating. Every YC batch dinner, he says, feels the way the Stockholm room felt during his talk.

    The Mystery of Serendipitous Meetings

    One specific benefit of density is serendipitous meetings, and Graham admits he does not fully understand why unplanned encounters outperform planned ones so dramatically. Biographies of accomplished people are dense with chance meetings that redirected entire lives. He offers three possible explanations. Maybe there are simply more unplanned meetings, so statistically the outliers will mostly be unplanned. Maybe planned meetings are too conservative because they require a stated reason in advance, which lops off the upside the same way deliberate startup idea hunts lop off the best ideas. Maybe unplanned conversations have built in selection. You can decide in the first few sentences whether to continue, so the surviving conversations are pre filtered for fit. Whatever the mechanism, big centers produce more of these high value encounters, and that alone is worth the move.

    Speed and the Investor Asymmetry

    Things move faster in big centers because better people are more confident and more decisive. They egg each other on. Ideas get acted on instead of half held. Graham notes that in villages around the world there are people who half had every famous idea and never moved on it, and now resent the founder who did.

    The starkest example is investor speed. Silicon Valley investors decide dramatically faster than European ones, partly because they are better and more confident and partly because competition forces it. An investor who correctly identifies a great opportunity faces a counterintuitive rule. The more right they are, the less time they can wait, because every other investor who meets that founder will reach the same conclusion. Yuri Sagalov is the canonical case. He invested in Max Levchin immediately on meeting him because he knew anyone else would do the same. Valley investors complain that valuations are too high and decisions too rushed, but they empirically outperform European investors anyway. The grumbling is noise.

    The Prophet at Home Effect

    An underrated benefit of leaving for the center is that it raises your standing at home. Graham quotes the line about no prophet in their own country and notes that investors outside Silicon Valley implicitly assume local startups are second rate. It is not a Swedish problem. It is universal. Leaving inverts the rule. Local investors automatically rate you higher because you have been somewhere they consider serious. Sometimes the mere announcement that you got into Y Combinator triggers the inversion. The Dropbox story is the cleanest illustration. A big Boston VC firm spent a year giving Drew Houston encouragement and advice but no money. The moment Sequoia took an interest in Silicon Valley, that same firm faxed Drew a term sheet with a blank valuation, willing to invest at any price. Drew went with Sequoia. Dropbox went public in 2018 as the first YC IPO.

    Big Pond, Visible Summit

    The deepest benefit of relocating is not what the center does for you but what it does to you. A big fish in a small pond cannot tell how big it actually is. A big fish in a big pond can. You can stand next to Brian Chesky or Sam Altman or, as the Stockholm audience just had, Max Levchin, and recognize that they are not a different species. You could do what they did, if you worked that hard. The catch, Graham emphasizes twice, is the if. Seeing a giant up close calibrates both the achievability of the summit and the cost of reaching it.

    He offers a Mount Olympus image. Moving to the mountain clears away the fog at the top. The summit is right there, quite high but no longer impossibly high. Ambitious people need a high but definite threshold. Visibility transforms a vague aspiration into a clear, hard, finite target.

    The Pay It Forward Culture

    The most surprising thing about Silicon Valley to outsiders is that people help you for no reason. The phrase sounds normal in the Valley and strange everywhere else, the way clean streets feel normal in Sweden but require explanation elsewhere. Graham asked a founder who recently moved from England what surprised him most. The answer was the helpfulness. Every conversation ended with what can I do to help you. The English founder noted that this was not English politeness, which is a different thing and arguably more pronounced.

    Graham traces the origin to economics. Silicon Valley is where nobodies become billionaires faster than anywhere else. Someone with a taste for being nice to nobodies, the kind of person who pets the nobody on the head rather than kicking it aside, was always going to end up with powerful friends in that environment. Whether the original behavior was calculated or not, it is reflexive now. The custom is 60 years old. Ron Conway is the purest expression. He helps everyone, does not track favors, does not remember most of them, and as a result operates at a scale that ledger keeping makes impossible. When many people behave that way at once, the conservation law for favors breaks down. The pie expands. Graham notes that moving to the Valley will change you in this same way, almost involuntarily.

    The Sweden Answer Is Inside the Founder Answer

    The pivot of the talk is that both questions have the same answer. The way Stockholm thrives as a startup hub is for Swedish founders to go to Silicon Valley and come back. That move helps Sweden in three concrete ways. The average quality of Swedish startups rises. Returning founders bring Valley money back with them. And they import Valley culture, which has been optimized over decades for startups and which is more compatible with Swedish culture than people assume. Sweden lacks the tall poppies dynamic, which it should drop anyway, and shares the high trust trait that the Valley runs on.

    The historical analogy is direct. In the late 1800s the Swedish government gave mathematicians fellowships conditional on leaving the country to study abroad. Boycotting Gutting to develop Swedish math would have been self defeating. The same logic applies to startups now.

    YC as the Optimal Vehicle

    Graham acknowledges he is talking his own book and says it anyway because he thinks it is true. The optimal way to go for a bit and come back is YC. YC is a deliberately engineered super valley inside the Valley, concentrating founder density, helpfulness, and investor speed into a four to six month container. If the Swedish government designed such a program from scratch it would look like YC, and YC costs the government nothing because Silicon Valley investors fund it. There is no licensing process. Founders just call the API.

    The Half As Many Unicorns Caveat

    The honest data point. Founders who go home after YC are only about half as likely to become unicorns as those who stay. Graham offers three reasons not to be discouraged. First, selection bias. The most confident and determined founders are also the ones willing to relocate, so the data is partly measuring those traits rather than the effect of geography. Second, the metric is valuation, not company performance. Bay Area companies simply raise at higher multiples. Third, half is still very good. A 500 million dollar company instead of a 1 billion dollar one is no real difference in practice, and in Swedish kroner you still cross the billionaire threshold.

    Money is not everything anyway. Once you have kids, where they grow up becomes the dominant decision, and that question has nothing to do with valuations.

    The Silicon Valley of Europe Is an Open Position

    Graham ends with the most ambitious frame. If Sweden transplants enough Valley culture, Stockholm could become the Silicon Valley of Europe. The job is unclaimed. There is no confident answer to where the European startup center is, the way nobody asks where the Silicon Valley of America is because the answer is obvious. Geographic size is a weaker constraint than people think. Mountain View was a backwater in 1955 when Shockley Semiconductor was founded there, and it remained the geometric center of Silicon Valley until activity shifted to San Francisco in 2012. The only real requirements are a place founders want to live and a critical mass of founders. Stockholm clearly clears the first bar. The second is impossible to measure until it is hit, and then it tips fast. Graham closes by suggesting Stockholm may already be closer than it looks.

    Thoughts

    The most useful idea in this talk is the inversion at the heart of it. Most advice about startup geography frames the choice as a tradeoff between leaving and staying, with leaving optimized for the founder and staying optimized for the country. Graham collapses the two. The country wins more when founders leave and come back than when founders stay out of loyalty. The brain drain framing assumes a fixed pool of talent that can only be in one place. The brain circulation framing, which is what Graham is actually describing, assumes that exposure compounds. A founder who has spent six months absorbing Valley density brings back something a founder who stayed home never had. The Swedish math fellowships from the 1800s are the deepest evidence here. A government that wanted strong domestic mathematicians did not try to build a wall around them. It paid them to leave.

    The serendipity argument is the part of the talk that should make planners uncomfortable, because it is essentially an admission that the highest leverage activity in a startup career cannot be scheduled. The three theories Graham offers are not mutually exclusive and the cumulative force of them is that any environment optimized for planned, calendared interaction is by definition lopping off its own upside. This has obvious implications beyond geography. Remote first cultures, calendar tetris, gated office access, and the whole apparatus that converts random encounters into booked meetings are all working against the mechanism Graham is describing. Whether that tradeoff is worth it for any given company is a separate question, but it is at minimum a tradeoff, not a free win.

    The pay it forward story is also more economically grounded than it usually gets credit for. Graham is careful to note that the helping behavior may have originated as a calculated bet on being kind to potential future billionaires, then ossified into reflex once enough generations practiced it. That is a more honest origin story than the usual quasi spiritual version. It also implies the culture can be transplanted, but only by recreating the conditions that originally produced it. You cannot just declare a pay it forward culture and have one. You need a place where nobodies actually do become billionaires often enough that helping them rationally pays off, then run that loop for 60 years. Most cities trying to engineer their way into being startup hubs skip past this part and wonder why the culture does not stick.

    Finally, the Mountain View in 1955 line is the underrated punch of the talk. People who write off their own city as too small or too peripheral to become anything usually have an idealized image of the current center as a place that was always obviously special. It was not. Shockley Semiconductor went into a strip of orchards. Whatever Stockholm or anywhere else looks like today, it looks more impressive than Mountain View did the year Silicon Valley was born.

    Watch the full Paul Graham talk from Stockholm on YouTube.

  • Andrej Karpathy on Vibe Coding vs Agentic Engineering: Why He Feels More Behind Than Ever in 2026

    Andrej Karpathy, co-founder of OpenAI, former head of AI at Tesla, and now founder of Eureka Labs, returned to Sequoia Capital’s AI Ascent 2026 stage for a wide-ranging conversation with partner Stephanie Zhan. One year after coining the term “vibe coding,” Karpathy unpacked what has changed, why he has never felt more behind as a programmer, and why the discipline emerging on top of vibe coding, which he calls agentic engineering, is the more serious craft worth learning right now.

    The conversation covered Software 3.0, the limits of verifiability, why LLMs are better understood as ghosts than animals, and why you can outsource your thinking but never your understanding. Below is a complete breakdown of the talk for anyone building, hiring, or learning in the agent era.

    TLDW

    Karpathy describes a sharp transition that happened in December 2025, when agentic coding tools crossed a threshold and code chunks just started coming out fine without correction. He frames the current moment as Software 3.0, where prompting an LLM is the new programming, and entire app categories are collapsing into a single model call. He distinguishes vibe coding (raising the floor for everyone) from agentic engineering (preserving the professional quality bar at much higher speed). Models remain jagged because they are trained on what labs choose to verify, so founders should look for valuable but neglected verifiable domains. Taste, judgment, oversight, and understanding remain uniquely human responsibilities, and tools that enhance understanding are the ones he is most excited about.

    Key Takeaways

    • December 2025 was a clear inflection point. Code chunks from agentic tools started arriving correct without edits, and Karpathy stopped correcting the system entirely.
    • Software 3.0 means programming has become prompting. The context window is your lever over the LLM interpreter, which performs computation in digital information space.
    • Open Code’s installer is a software 3.0 example. Instead of a complex shell script, you copy paste a block of text to your agent, and the agent figures out your environment.
    • The Menu Gen anecdote illustrates how entire apps can become spurious. What used to require OCR, image generation, and a hosted Vercell app can now be a single Gemini plus Nano Banana prompt.
    • Vibe coding raises the floor. Agentic engineering preserves the professional ceiling. The two are different disciplines.
    • The 10x engineer multiplier is now far higher than 10x for people who are good at agentic engineering.
    • Hiring processes have not caught up. Puzzle interviews are the old paradigm. New evaluations should look like building a full Twitter clone for agents and surviving simulated red team attacks from other agents.
    • Models are jagged because reinforcement learning rewards what is verifiable, and labs choose which verifiable domains to invest in. Strawberry letter counts and the 50 meter car wash question show how state-of-the-art models can refactor 100,000 line codebases yet fail at trivial reasoning.
    • If you are in a verifiable setting, you can run your own fine tuning, build RL environments, and benefit even when the labs are not focused on your domain.
    • LLMs are ghosts, not animals. They are statistical simulations summoned from pre training and shaped by RL appendages, not creatures with curiosity or motivation. Yelling at them does not help.
    • Taste, aesthetics, spec design, and oversight remain human jobs. Models still produce bloated, copy paste heavy code with brittle abstractions.
    • Documentation is still written for humans. Agent native infrastructure, where docs are explicitly designed to be copy pasted into an agent, is a major opportunity.
    • The future likely involves agent representation for people and organizations, with agents talking to other agents to coordinate meetings and tasks.
    • You can outsource your thinking but not your understanding. Tools that help humans understand information faster are uniquely valuable.

    Detailed Summary

    Why Karpathy Feels More Behind Than Ever

    Karpathy opens by describing how he has been using agentic coding tools for over a year. For most of that period, the experience was mixed. The tools could write chunks of code, but they often required edits and supervision. December 2025 changed everything. With more time during a holiday break and the release of newer models, Karpathy noticed that the chunks just came out fine. He kept asking for more. He cannot remember the last time he had to correct the agent. He started trusting the system, and what followed was a cascade of side projects.

    He wants to stress that anyone whose model of AI was formed by ChatGPT in early 2025 needs to look again. The agentic coherent workflow that genuinely works is a fundamentally different experience, and the transition was stark.

    Software 3.0 Explained

    The Software 1.0 paradigm was writing explicit code. Software 2.0 was programming by curating datasets and training neural networks. Software 3.0 is programming by prompting. When you train a GPT class model on a sufficiently large set of tasks, the model implicitly learns to multitask everything in the data. The result is a programmable computer where the context window is your interface, and the LLM is the interpreter performing computation in digital information space.

    Karpathy gives two concrete examples. The first is Open Code’s installer. Normally a shell script handles installation across many platforms, and these scripts balloon in complexity. Open Code instead provides a block of text you copy paste to your agent. The agent reads your environment, follows instructions, debugs in a loop, and gets things working. You no longer specify every detail. The agent supplies its own intelligence.

    The Menu Gen Story

    The second example is Karpathy’s Menu Gen project. He built an app that takes a photo of a restaurant menu, OCRs the items, generates pictures for each dish, and renders the enhanced menu. The app runs on Vercell and chains together multiple services. Then he saw a software 3.0 alternative. You take a photo, give it to Gemini, and ask it to use Nano Banana to overlay generated images onto the menu. The model returns a single image with everything rendered. The entire app he built is now spurious. The neural network does the work. The prompt is the photo. The output is the photo. There is no app between them.

    Karpathy uses this to argue that founders should not just think of AI as a speedup of existing patterns. Entirely new things become possible. His example is LLM driven knowledge bases that compile a wiki for an organization from raw documents. That is not a faster version of older code. It is a new capability with no prior equivalent.

    What Will Look Obvious in Hindsight

    Stephanie Zhan asks what the equivalent of building websites in the 1990s or mobile apps in the 2010s looks like today. Karpathy speculates about completely neural computers. Imagine a device that takes raw video and audio as input, runs a neural net as the host process, and uses diffusion to render a unique UI for each moment. He notes that early computing in the 1950s and 60s was undecided between calculator like and neural net like architectures. We went down the calculator path. He thinks the relationship may eventually flip, with neural networks becoming the host and CPUs becoming co processors used for deterministic appendages.

    Verifiability and Jagged Intelligence

    Karpathy spent significant writing time on verifiability. Classical computers automate what you can specify in code. The current generation of LLMs automates what you can verify. Frontier labs train models inside giant reinforcement learning environments, so the models peak in capability where verification rewards are strong, especially math and code. They stagnate or get rough around the edges elsewhere.

    This explains the jagged intelligence puzzle. The classic example was counting letters in strawberry. The newer one Karpathy offers: a state of the art model will refactor a 100,000 line codebase or find zero day vulnerabilities, then tell you to walk to a car wash 50 meters away because it is so close. The two coexisting capabilities should be jarring. They reveal that you must stay in the loop, treat models as tools, and understand which RL circuits your task lands in.

    He also points out that data distribution choices matter. The jump in chess capability from GPT 3.5 to GPT 4 came largely because someone at OpenAI added a huge amount of chess data to pre training. Whatever ends up in the mix gets disproportionately good. You are at the mercy of what labs prioritize, and you have to explore the model the labs hand you because there is no manual.

    Founder Advice in a Lab Dominated World

    Asked what founders should do given that labs are racing toward escape velocity in obvious verifiable domains, Karpathy points back to verifiability itself. If your domain is verifiable but currently neglected, you can build RL environments and run your own fine tuning. The technology works. Pull the lever with diverse RL environments and a fine tuning framework, and you get something useful. He hints there is one specific domain he finds undervalued but declines to name it on stage.

    On the question of what is automatable only from a distance, Karpathy says almost everything can ultimately be made verifiable. Even writing can be assessed by councils of LLM judges. The differences are in difficulty, not in possibility.

    From Vibe Coding to Agentic Engineering

    Vibe coding raises the floor. Anyone can build something. Agentic engineering preserves the professional quality bar that existed before. You are still responsible for your software. You are still not allowed to ship vulnerabilities. The question is how you go faster without sacrificing standards. Karpathy calls it an engineering discipline because coordinating spiky, stochastic agents to maintain quality at speed requires real skill.

    The ceiling on agentic engineering capability is very high. The old idea of a 10x engineer is now an understatement. People who are good at this peak far above 10x.

    What Mediocre Versus AI Native Looks Like

    Karpathy compares this to how different generations use ChatGPT. The difference between a mediocre and an AI native engineer using Claude Code, Codex, or Open Code is investment in setup and full use of available features. The same way previous generations of engineers got the most out of Vim or VSCode, today’s strong engineers tune their agentic environments deeply.

    He thinks hiring processes have not caught up. Most companies still hand out puzzles. The new test should look like asking a candidate to build a full Twitter clone for agents, make it secure, simulate user activity with agents, and then run multiple Codex 5.4x high instances trying to break it. The candidate’s system should hold up.

    What Humans Still Own

    Agents are intern level entities right now. Humans are responsible for aesthetics, judgment, taste, and oversight. Karpathy describes a Menu Gen bug where the agent tried to associate Stripe purchases with Google accounts using email addresses as the key, instead of a persistent user ID. Email addresses can differ between Stripe and Google accounts. This kind of specification level mistake is exactly what humans must catch.

    He works with agents to design detailed specs and treats those as documentation. The agent fills in the implementation. He has stopped memorizing API details for things like NumPy axis arguments or PyTorch reshape versus permute. The intern handles recall. Humans handle architecture, design, and the right questions.

    Reading the actual code agents produce can still cause heart attacks. It is bloated, full of copy paste, riddled with awkward and brittle abstractions. His Micro GPT project, an attempt to simplify LLM training to its bare essence, was nearly impossible to drive through agents. The models hate simplification. That capability sits outside their RL circuits. Nothing is fundamentally preventing this from improving. The labs simply have not invested.

    Animals Versus Ghosts

    Karpathy returns to his framing that we are not building animals, we are summoning ghosts. Animal intelligence comes from evolution and is shaped by intrinsic motivation, fun, curiosity, and empowerment. LLMs are statistical simulation circuits where pre training is the substrate and RL is bolted on as appendages. They are jagged. They do not respond to being yelled at. They have no real curiosity. The ghost framing is partly philosophical, but it changes how you approach them. You stay suspicious. You explore. You do not assume the system you used yesterday will behave the same on a new task.

    Agent Native Infrastructure

    Most software, frameworks, libraries, and documentation are still written for humans. Karpathy’s pet peeve is being told to do something instead of being given a block of text to copy paste to his agent. He wants agent first infrastructure. The Menu Gen project’s hardest part was not writing code. It was deploying on Vercell, configuring DNS, navigating service settings, and stringing together integrations. He wants to give a single prompt and have the entire thing deployed without touching anything.

    Long term he expects agent representation for individuals and organizations. His agent will negotiate meeting details with your agent. The world becomes one of sensors, actuators, and agent native data structures legible to LLMs.

    Education and What Still Matters

    The most striking line of the conversation comes near the end. Karpathy quotes a tweet that shaped his thinking: you can outsource your thinking but you cannot outsource your understanding. Information still has to make it into your brain. You still need to know what you are building and why. You cannot direct agents well if you do not understand the system.

    This is part of why he is so excited about LLM driven knowledge bases. Every time he reads an article, his personal wiki absorbs it, and he can query it from new angles. Every projection onto the same information yields new insight. Tools that enhance human understanding are uniquely valuable because LLMs do not excel at understanding. That bottleneck is yours to manage.

    Thoughts

    The most useful frame in this talk is the distinction between vibe coding and agentic engineering. It clarifies what has been muddled for the past year. Vibe coding is about access. Anyone can produce something. Agentic engineering is about discipline. You preserve the standards that made software trustworthy in the first place, while moving at speeds that would have seemed absurd two years ago. These are not the same activity, and conflating them is part of why so many shipped products feel half built.

    The Menu Gen anecdote is the kind of story that should make every solo developer pause. If a single Gemini plus Nano Banana prompt can replace a multi service Vercell deployed app, the question for any builder becomes how much of what you are working on right now is going to be made spurious by the next model release. The honest answer is probably more than you want to admit. The defensive posture is not building thicker apps. It is choosing problems where the model alone is not enough, where taste, distribution, infrastructure, or specific verifiable RL environments give you something the next model cannot collapse into a prompt.

    The verifiability lens is also unusually practical. If you are a solo builder, the question shifts from what is possible to what is verifiable but neglected. The labs will eat the obvious verifiable domains because that is how their RL pipelines are set up. The opportunity is in domains where verification is possible but the labs have not yet invested. That is a much more concrete strategic filter than vague intuitions about defensibility.

    The car wash example is going to stick. State of the art models can refactor enormous codebases and still tell you to walk somewhere a sane person would drive. That is the lived reality of jagged intelligence, and it argues strongly for staying in the loop on real decisions rather than handing off everything to agents. The agents are excellent fillers of blanks. They are not yet trustworthy specifiers of the spec.

    Finally, the line about outsourcing thinking but not understanding is worth taping above the desk. The bottleneck is no longer typing speed, syntax recall, or even API knowledge. It is whether the human in the loop actually understands the system being built. Tools that genuinely improve human understanding, including personal knowledge bases that re project information through different prompts, are likely the most undervalued category of products being built right now. The opportunity is not just in agents. It is in the cognitive scaffolding that makes humans good directors of agents.