Andrej Karpathy
Founder, Eureka Labs · Co-founder of OpenAI (early) · Former Director of AI, Tesla (Autopilot lead, 2017–2022) · PhD student of Geoff Hinton; long-time deep learning practitioner
“the Decade of Agents, that's actually a reaction to an existing pre. Existing quote... I think I was triggered by that because I feel like there's some over predictions going on in the industry. And in my mind this is really a lot more accurately described as the Decade of Agents.”
“When would you prefer to have an agent like Cloud or Codex do that work? Currently, of course, they can't. What would it take for them to be able to do that?... they just don't work. So they don't have enough intelligence, they're not multimodal enough. They can't do computer use and all this kind of stuff... They don't have continual learning. You can't just tell them something and they'll remember it, and they're just cognitively lacking, and it's just not working. And I just think that it will take about a decade to work through all of those issues.”
“We're not actually building animals, we're building ghosts. These like sort of ethereal spirit entities because they're fully digital and they're kind of like mimicking humans. And it's a different kind of intelligence... we're not doing training by evolution. We're doing training by basically imitation of humans and the data that they've put on the Internet.”
“I think you're getting at some of my, like why my timelines are a bit longer. You're right. I think yeah, they're not very good at code that has never been written before. Maybe is like one way to put it, which is like what we're trying to achieve when we're building these models.”
“the agents are actually pretty good. For example, if you're doing boilerplate stuff, boilerplate code that's just copy paste stuff, they're very good at that. They're very good at stuff that occurs very often on the Internet because there's lots of examples of it in the training sets of these models... [nanochat] is not an example of those because it's a fairly unique repository... they keep, they keep thinking I'm writing normal code and I'm not.”
“there's like three major classes of how people interact with code right now. Some people completely reject all of LLMs and they are just writing by scratch. I think this is probably not the right thing to do anymore. The intermediate part, which is where I am is you still write a lot of things from scratch, but you use the autocomplete that's basically available now from these models... And then there's the vibe coding. Hi, please implement this or that, enter and then let the model do it. And that's the agents.”
“reinforcement learning is terrible. It just so happens that everything that we had before is much worse... Literally, what reinforcement learning does is it goes to the ones that worked really well, and every single thing you did along the way, every single token gets upweighted of, like, do more of this... you're sucking supervision through a straw because you've done all this work that could be a minute of rollout, and you're, like, sucking the bits of supervision of the final reward signal through a straw.”
“anytime you use an LLM to assign a reward, those LLMs are giant things with billions of parameters and they're gameable. And if you're reinforcement learning with respect to them, you will find adversarial examples for your LLM judges. Almost guaranteed... it turns out that duh duh duh duh is an adversarial example for the model and it assigns 100% probability to it.”
“every single nine is the same amount of work. So when you get a demo and something works 90% of the time, that's just the first nine and then you need the second nine and third nine, fourth nine, fifth nine. And while I was at Tesla for was it five years or so, I think we went through maybe three nines or two nines... software is a critical safety domain, unless you're doing by coding, which is all nice and fun and so on. And so that's why I think this also enforced my timelines from that perspective.”
“what I think we have to do kind of going forward, and this would be part of the research paradigms, is actually think we need to figure out ways to remove some of the knowledge and to keep what I call this cognitive core as this intelligent entity that is kind of stripped from knowledge, but contains the algorithms and contains the magic of intelligence and problem solving and the strategies of it and all this kind of stuff.”
“you've talked about how you were at Tesla leading self driving from 2017 to 2022 and then you firsthand saw this progress from we went from cool demos to now thousands of cars out there actually autonomously doing drives. Why did that take a decade?... I would say one thing I will almost instantly also push back on is this is not even near done so in a bunch of ways that I'm going to get to.”
“it's a march of nines. And every single nine is a constant amount of work. So every single nine is the same amount of work. So when you get a demo and something works 90% of the time, that's just the first nine and then you need the second nine and third nine, fourth nine, fifth nine. And while I was at Tesla for was it five years or so, I think we went through maybe three nines or two nines.”
“when I was joining Tesla I had a very early demo of a Waymo and it basically gave me a perfect drive in 2014 or something like that. So perfect. Waymo Drive a decade ago took us around Palo Alto and so on, because I had a friend who worked either and I thought it was like very close and then still took a long time.”
“self driving cars are nowhere near done still. So the deployments still are pretty minimal. So even Waymo and so on has very few cars. And they're doing that, roughly speaking, because they're not economical, because they've built something that lives in the future. And so they had to pull back future, but they had to make it uneconomical. So they have all these, there's all these costs, not just marginal costs for those cars and their operation and maintenance, but also the capex of the entire thing. So making it economical is still going to be a slog, I think, for them.”
“when you look at these cars and there's no one driving, I also think it's a little bit deceiving because there are actually very elaborate teleoperation centers of people actually kind of like in a loop with these cars. And I don't have the full extent of it, but I think there's more human in the loop that you might expect. And there's people somewhere out there basically beaming in from the sky. And I don't actually know they're fully in the loop with the driving. I think some of the times they are are, but they're certainly involved and there are people and in some sense we haven't actually removed the person, we've moved them to somewhere we can't see them.”
“Waymo can't go to all the different parts of the city. My suspicion is it's like parts of city where you don't get good signals anyway. So basically I don't actually know anything about the stack. I mean, I'm just making up stuff... I just think people again are sometimes a little bit too naive about some of the progress and I still think there's a huge amount of work and I think Tesla took in my mind a lot more scalable approach and I think the team is doing extremely well and it's going to. And I'm kind of like on the record for predicting how this thing will go, which is like way more early start because you can package up so many sensors. But I do think Tesla is taking the more scalable strategy and it's going to look a lot more like that.”
“for some kinds of tasks and jobs and so on, there's a very large demo to product gap where the demo is very easy, but the product is very hard. And it's especially the case in cases like self driving where the cost of failure is too high... I'm very unimpressed by demos. So whenever I see demos of anything, I'm extremely unimpressed by that. It works better if you can. If it's a demo that someone cooked up and is just showing you its worst, if you can interact with it, it's a bit better.”
“It's business as usual because we're in an intelligence explosion already and have been for decades. Everything is gradual visually being automated has been for hundreds of years.”
“I do feel like I have a hard time differentiating where AI begins and stops because I do see AI as fundamentally an extension of computing in some pretty fundamental way. And I feel like I see a continuum of this kind of recursive self improvement or of speeding up programmers all the way from the beginning... We're not writing the assembly code because we have compilers, right? Like compilers will take my high level language in C and write the assembly code. So we're abstracting ourselves very, very slowly. And there's this What I call autonomy slider of like more and more stuff is automated of the stuff that can be automated at any point in time. And we're doing a bit less and less on raising ourselves in the labor abstraction over the automation.”
“If you have a perfect AI tutor, maybe you can get extremely far. The geniuses of today are barely scratching the surface of what a human mind can do, I think.”
“computing is labor. Computing was labor. Computers, like a lot of jobs disappeared because computers are automating a bunch of digital information processing that you now don't need a human for. And so computers are labor and that has played out out. And self driving as an example is also like computers doing labor. So I guess that's already been playing out. So still business as usual.”
“in December is when it really just something flipped where I kind of went from 80, 20 of to like 2080 of writing code by myself versus just delegating to agents. And I don't even think it's 2080 by now. I think it's a lot more than that. I don't think I've typed like a line of code probably since December, basically, which is like an extremely large change... if you just find a random Software engineer or something like that at their desk. And what they're doing, they're default workflow of building software is completely different as of basically December.”
“How can I have not just a single session of clock code or codecs or some of these agent harnesses? How can I have more of them? How can I do that appropriately?... here's a new functionality and delegate it to agent one. Here's a new functionality that's not going to interfere with the other one, give it agent two and then try to review their work as best as you can, depending on how much you care about that code. What are these macro actions that I can manipulate my software repository by? And another agent is doing some research, another agent is writing code, another one is coming up with a plan for some new implementation.”
“It's a skill issue, which is very empowering because... it's not that the capability is not there, it's that you just haven't found a way to string it together of what's available. Like, I just don't. I didn't give good enough Instructions in the AgentsMD file or whatever it may be. I don't have a nice enough memory tool that I put in there or something like that. So it all kind of feels like skill issue when it doesn't work.”
“I simultaneously feel like I'm talking to an extremely brilliant PhD student who's been like a systems programmer for their entire life and a 10 year old. And it's so weird because humans, I feel like they're a lot more coupled... humans have a lot more jaggedness, although they definitely have some. But humans have a lot less of that kind of jaggedness, although they definitely have some. But the agents have a lot more jaggedness where sometimes I ask for functionality and it comes back with something that's just totally wrong and then we get into loops that are totally wrong.”
“these models are trained via reinforcement learning. So they're actually struggling with the exact same thing we just talked about, which is the labs can improve the models in anything that is verifiable, but that has rewards. So did you write the program correctly and do the unit test checkout? Yes or no. But some of the things where they're struggling is like, for example, I think they have a tough time with nuance of maybe what I had in mind or what I intended and when to ask clarifying questions. Anything that feels softer is worse. And so you're kind of like you're either on rails and you're part of the super intelligence circuits, or you're not on rails and you're outside of the verifiable domains and suddenly everything kind of just meanders.”
“this is the joke you would get, like, three or four years ago, and this is the joke you still get today... even though the models have improved tremendously, and if you give them an agentic task, they will just go for hours and move mountains for you. And then you ask for, like, a joke, and it has a stupid joke, a crappy joke from five years ago, and it's because it's outside of the rl. It's outside of the reinforcement learning. It's outside of what's being improved.”
“I had a tweet earlier where I kind of like said something along the lines of to get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You can't be there to prompt the next thing. You need to take yourself outside. You have to arrange things such that they're completely autonomous... auto research is just, yeah, here's an objective, here's a metric, here's your boundaries of what you can and cannot do. And go.”
“I let autoresearch go for overnight and it came back with tunings that I didn't see. And yeah, I did forget the weight decay on the value embeddings and my atom betas were not sufficiently tuned. And these things jointly interact. So once you tune one thing, the other things have to potentially change too. I shouldn't be a bottleneck, I shouldn't be running these hyperparameters search optimizations. I shouldn't be looking at the results.”
“if you have a bunch of nodes of parallelization available to you, then it's very easy to just have multiple auto researchers talking through a common system or something like that. What I was more interested in is how you can have an untrusted pool of workers out there on the Internet... a swarm of agents on the Internet could collaborate to improve LLMs and could potentially even run circles around Frontier Labs, who knows? Yeah, maybe that's even possible. Frontier Labs have a huge amount of trusted compute, but the Earth is much bigger and has huge amount of untrusted compute.”
“I think currently my impression is the labs are trying to have a single sort of monoculture of a model that is arbitrarily intelligent in all these different domains and they just stuff into the parameters. I do think we should expect more speciation in the intelligences. The animal kingdom is extremely diverse in the brains that exist, and there's lots of different niches of nature and some animals have overdeveloped visual cortex or other kind of parts.”
“I do have a cautiously optimistic view of this in software engineering, where it does seem to me like the demand for software will be extremely large and it's just become a lot cheaper. So I do think that for quite some time, it's very hard to forecast. But it does seem to me like right now, at least locally, there's going to be more demand for software because software is amazing... if the barrier comes down, then actually you have the Jevons paradox, which is like, you know, actually the demand for software actually goes up, up. It's cheaper and there's more powerful. The classical example of this always is the ATMs and the bank tellers, because there was a lot of fear that ATMs and computers basically would displace tellers. But what happened is they made the cost of operation of a bank branch much cheaper as there were more bank branches, so there were more tellers.”
“I did spend a bunch of time going around opening eye and I was like, you guys realize if we're successful. Like we're all out of job. Like we're just building automation for Sam or something like that. Or the board, I'm not sure, but they're just building this automation for the board or the CEO or something like that. And we're all out of our job and maybe contributing on the sides.”
“if the barrier comes down, then actually you have the Jevons paradox, which is like, you know, actually the demand for software actually goes up, up. It's cheaper and there's more powerful. The classical example of this always is the ATMs and the bank tellers, because there was a lot of fear that ATMs and computers basically would displace tellers. But what happened is they made the cost of operation of a bank branch much cheaper as there were more bank branches, so there were more tellers is like the canonical example people cite. But basically it's just Jevons paradox. Something becomes cheaper, so there's a lot of unlocked demand for it.”
“these jobs are bundles of tasks and some of these tasks can go a lot faster. And so people should think of it as primarily a tool that it is right now. And I think the long term future of that is uncertain. Yeah, it's kind of really hard to forecast, to be honest. And I'm not professionally doing that really. And I think it's the job of economists to do properly... just being, you know, just trying to keep up with it is like the first thing.”
“Yeah, what is it limited by? Just, I think, everything. Like so many things, even if they don't work, I think to a large extent you feel like it's a skill issue. It's not that the capability is not there, it's that you just haven't found a way to string it together of what's available. Like, I just don't. I didn't give good enough Instructions in the AgentsMD file or whatever it may be.”
“[Karpathy joining Anthropic to lead a new pre-training team focused on recursive self-improvement]: he led the self driving team at Tesla also... he was probably the first person that really commercialized the Richard Sutton bitter lesson essay when he was leading FSD at Tesla, which was really about the brute force computation... he's been at the wave upon wave of AI... he's one of these really curious people that can be sent off and they'll just go and invent new things. And I think this idea of recursive self learning puts these models on a combination of overdrive and autopilot. — [[chamath-palihapitiya]] in [[2026-05-22-podcast-all-in-podcast-spacex-s-2t-case-nvidia-s-shock-selloff-america]]”
Andrej Karpathy
One-line summary: Frontier AI practitioner with operator-grade perspective spanning Tesla Autopilot, OpenAI, and deep learning research. Tracked here for capability-timeline framings and AI-agent productivity claims.
What they're known for
Brief factual context — fill in.
Why they matter to artificial-intelligence
Why this person's claims are tracked here — fill in.
Said
Speaker-attributed claims extracted from diarized sources. Each bullet mirrors one entry in quotes: frontmatter — keep them in sync.
-
On agi-timeline-decade-of-agents:
"the Decade of Agents, that's actually a reaction to an existing pre. Existing quote... I think I was triggered by that because I feel like there's some over predictions going on in the industry. And in my mind this is really a lot more accurately described as the Decade of Agents." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents:
"When would you prefer to have an agent like Cloud or Codex do that work? Currently, of course, they can't. What would it take for them to be able to do that?... they just don't work. So they don't have enough intelligence, they're not multimodal enough. They can't do computer use and all this kind of stuff... They don't have continual learning. You can't just tell them something and they'll remember it, and they're just cognitively lacking, and it's just not working. And I just think that it will take about a decade to work through all of those issues." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents:
"We're not actually building animals, we're building ghosts. These like sort of ethereal spirit entities because they're fully digital and they're kind of like mimicking humans. And it's a different kind of intelligence... we're not doing training by evolution. We're doing training by basically imitation of humans and the data that they've put on the Internet." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On ai-coding-agent-asymmetry-on-novel-code, ai-coding-productivity-paradox, agi-timeline-decade-of-agents:
"I think you're getting at some of my, like why my timelines are a bit longer. You're right. I think yeah, they're not very good at code that has never been written before. Maybe is like one way to put it, which is like what we're trying to achieve when we're building these models." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On ai-coding-agent-asymmetry-on-novel-code, ai-coding-productivity-paradox:
"the agents are actually pretty good. For example, if you're doing boilerplate stuff, boilerplate code that's just copy paste stuff, they're very good at that. They're very good at stuff that occurs very often on the Internet because there's lots of examples of it in the training sets of these models... [nanochat] is not an example of those because it's a fairly unique repository... they keep, they keep thinking I'm writing normal code and I'm not." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On ai-coding-productivity-paradox, ai-coding-tool-landscape-2026:
"there's like three major classes of how people interact with code right now. Some people completely reject all of LLMs and they are just writing by scratch. I think this is probably not the right thing to do anymore. The intermediate part, which is where I am is you still write a lot of things from scratch, but you use the autocomplete that's basically available now from these models... And then there's the vibe coding. Hi, please implement this or that, enter and then let the model do it. And that's the agents." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents:
"reinforcement learning is terrible. It just so happens that everything that we had before is much worse... Literally, what reinforcement learning does is it goes to the ones that worked really well, and every single thing you did along the way, every single token gets upweighted of, like, do more of this... you're sucking supervision through a straw because you've done all this work that could be a minute of rollout, and you're, like, sucking the bits of supervision of the final reward signal through a straw." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents:
"anytime you use an LLM to assign a reward, those LLMs are giant things with billions of parameters and they're gameable. And if you're reinforcement learning with respect to them, you will find adversarial examples for your LLM judges. Almost guaranteed... it turns out that duh duh duh duh is an adversarial example for the model and it assigns 100% probability to it." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents, ai-coding-agent-asymmetry-on-novel-code:
"every single nine is the same amount of work. So when you get a demo and something works 90% of the time, that's just the first nine and then you need the second nine and third nine, fourth nine, fifth nine. And while I was at Tesla for was it five years or so, I think we went through maybe three nines or two nines... software is a critical safety domain, unless you're doing by coding, which is all nice and fun and so on. And so that's why I think this also enforced my timelines from that perspective." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents:
"what I think we have to do kind of going forward, and this would be part of the research paradigms, is actually think we need to figure out ways to remove some of the knowledge and to keep what I call this cognitive core as this intelligent entity that is kind of stripped from knowledge, but contains the algorithms and contains the magic of intelligence and problem solving and the strategies of it and all this kind of stuff." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On demo-to-product-gap-march-of-nines, tesla-fsd:
"you've talked about how you were at Tesla leading self driving from 2017 to 2022 and then you firsthand saw this progress from we went from cool demos to now thousands of cars out there actually autonomously doing drives. Why did that take a decade?... I would say one thing I will almost instantly also push back on is this is not even near done so in a bunch of ways that I'm going to get to." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On demo-to-product-gap-march-of-nines, tesla-fsd:
"it's a march of nines. And every single nine is a constant amount of work. So every single nine is the same amount of work. So when you get a demo and something works 90% of the time, that's just the first nine and then you need the second nine and third nine, fourth nine, fifth nine. And while I was at Tesla for was it five years or so, I think we went through maybe three nines or two nines." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On waymo, demo-to-product-gap-march-of-nines:
"when I was joining Tesla I had a very early demo of a Waymo and it basically gave me a perfect drive in 2014 or something like that. So perfect. Waymo Drive a decade ago took us around Palo Alto and so on, because I had a friend who worked either and I thought it was like very close and then still took a long time." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On waymo:
"self driving cars are nowhere near done still. So the deployments still are pretty minimal. So even Waymo and so on has very few cars. And they're doing that, roughly speaking, because they're not economical, because they've built something that lives in the future. And so they had to pull back future, but they had to make it uneconomical. So they have all these, there's all these costs, not just marginal costs for those cars and their operation and maintenance, but also the capex of the entire thing. So making it economical is still going to be a slog, I think, for them." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On waymo, does-remote-supervised-robotaxi-qualify-as-driverless:
"when you look at these cars and there's no one driving, I also think it's a little bit deceiving because there are actually very elaborate teleoperation centers of people actually kind of like in a loop with these cars. And I don't have the full extent of it, but I think there's more human in the loop that you might expect. And there's people somewhere out there basically beaming in from the sky. And I don't actually know they're fully in the loop with the driving. I think some of the times they are are, but they're certainly involved and there are people and in some sense we haven't actually removed the person, we've moved them to somewhere we can't see them." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On tesla-fsd, vision-only-vs-sensor-fusion:
"Waymo can't go to all the different parts of the city. My suspicion is it's like parts of city where you don't get good signals anyway. So basically I don't actually know anything about the stack. I mean, I'm just making up stuff... I just think people again are sometimes a little bit too naive about some of the progress and I still think there's a huge amount of work and I think Tesla took in my mind a lot more scalable approach and I think the team is doing extremely well and it's going to. And I'm kind of like on the record for predicting how this thing will go, which is like way more early start because you can package up so many sensors. But I do think Tesla is taking the more scalable strategy and it's going to look a lot more like that." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On demo-to-product-gap-march-of-nines:
"for some kinds of tasks and jobs and so on, there's a very large demo to product gap where the demo is very easy, but the product is very hard. And it's especially the case in cases like self driving where the cost of failure is too high... I'm very unimpressed by demos. So whenever I see demos of anything, I'm extremely unimpressed by that. It works better if you can. If it's a demo that someone cooked up and is just showing you its worst, if you can interact with it, it's a bit better." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
"It's business as usual because we're in an intelligence explosion already and have been for decades. Everything is gradual visually being automated has been for hundreds of years." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On coder-to-builder-transition, ai-vampire-pattern:
"I do feel like I have a hard time differentiating where AI begins and stops because I do see AI as fundamentally an extension of computing in some pretty fundamental way. And I feel like I see a continuum of this kind of recursive self improvement or of speeding up programmers all the way from the beginning... We're not writing the assembly code because we have compilers, right? Like compilers will take my high level language in C and write the assembly code. So we're abstracting ourselves very, very slowly. And there's this What I call autonomy slider of like more and more stuff is automated of the stuff that can be automated at any point in time. And we're doing a bit less and less on raising ourselves in the labor abstraction over the automation." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
"If you have a perfect AI tutor, maybe you can get extremely far. The geniuses of today are barely scratching the surface of what a human mind can do, I think." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
"computing is labor. Computing was labor. Computers, like a lot of jobs disappeared because computers are automating a bunch of digital information processing that you now don't need a human for. And so computers are labor and that has played out out. And self driving as an example is also like computers doing labor. So I guess that's already been playing out. So still business as usual." — 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts (2025-10-17)
-
On agi-timeline-decade-of-agents, ai-coding-agent-asymmetry-on-novel-code, ai-coding-productivity-paradox:
"in December is when it really just something flipped where I kind of went from 80, 20 of to like 2080 of writing code by myself versus just delegating to agents. And I don't even think it's 2080 by now. I think it's a lot more than that. I don't think I've typed like a line of code probably since December, basically, which is like an extremely large change... if you just find a random Software engineer or something like that at their desk. And what they're doing, they're default workflow of building software is completely different as of basically December." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On ai-coding-productivity-paradox, ai-coding-tool-landscape-2026:
"How can I have not just a single session of clock code or codecs or some of these agent harnesses? How can I have more of them? How can I do that appropriately?... here's a new functionality and delegate it to agent one. Here's a new functionality that's not going to interfere with the other one, give it agent two and then try to review their work as best as you can, depending on how much you care about that code. What are these macro actions that I can manipulate my software repository by? And another agent is doing some research, another agent is writing code, another one is coming up with a plan for some new implementation." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On agi-timeline-decade-of-agents, ai-coding-agent-asymmetry-on-novel-code:
"It's a skill issue, which is very empowering because... it's not that the capability is not there, it's that you just haven't found a way to string it together of what's available. Like, I just don't. I didn't give good enough Instructions in the AgentsMD file or whatever it may be. I don't have a nice enough memory tool that I put in there or something like that. So it all kind of feels like skill issue when it doesn't work." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On agi-timeline-decade-of-agents, llm-capability-jaggedness:
"I simultaneously feel like I'm talking to an extremely brilliant PhD student who's been like a systems programmer for their entire life and a 10 year old. And it's so weird because humans, I feel like they're a lot more coupled... humans have a lot more jaggedness, although they definitely have some. But humans have a lot less of that kind of jaggedness, although they definitely have some. But the agents have a lot more jaggedness where sometimes I ask for functionality and it comes back with something that's just totally wrong and then we get into loops that are totally wrong." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On agi-timeline-decade-of-agents, llm-capability-jaggedness:
"these models are trained via reinforcement learning. So they're actually struggling with the exact same thing we just talked about, which is the labs can improve the models in anything that is verifiable, but that has rewards. So did you write the program correctly and do the unit test checkout? Yes or no. But some of the things where they're struggling is like, for example, I think they have a tough time with nuance of maybe what I had in mind or what I intended and when to ask clarifying questions. Anything that feels softer is worse. And so you're kind of like you're either on rails and you're part of the super intelligence circuits, or you're not on rails and you're outside of the verifiable domains and suddenly everything kind of just meanders." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
"this is the joke you would get, like, three or four years ago, and this is the joke you still get today... even though the models have improved tremendously, and if you give them an agentic task, they will just go for hours and move mountains for you. And then you ask for, like, a joke, and it has a stupid joke, a crappy joke from five years ago, and it's because it's outside of the rl. It's outside of the reinforcement learning. It's outside of what's being improved." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On autoresearch-recursive-self-improvement:
"I had a tweet earlier where I kind of like said something along the lines of to get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You can't be there to prompt the next thing. You need to take yourself outside. You have to arrange things such that they're completely autonomous... auto research is just, yeah, here's an objective, here's a metric, here's your boundaries of what you can and cannot do. And go." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On autoresearch-recursive-self-improvement:
"I let autoresearch go for overnight and it came back with tunings that I didn't see. And yeah, I did forget the weight decay on the value embeddings and my atom betas were not sufficiently tuned. And these things jointly interact. So once you tune one thing, the other things have to potentially change too. I shouldn't be a bottleneck, I shouldn't be running these hyperparameters search optimizations. I shouldn't be looking at the results." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On autoresearch-recursive-self-improvement:
"if you have a bunch of nodes of parallelization available to you, then it's very easy to just have multiple auto researchers talking through a common system or something like that. What I was more interested in is how you can have an untrusted pool of workers out there on the Internet... a swarm of agents on the Internet could collaborate to improve LLMs and could potentially even run circles around Frontier Labs, who knows? Yeah, maybe that's even possible. Frontier Labs have a huge amount of trusted compute, but the Earth is much bigger and has huge amount of untrusted compute." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
"I think currently my impression is the labs are trying to have a single sort of monoculture of a model that is arbitrarily intelligent in all these different domains and they just stuff into the parameters. I do think we should expect more speciation in the intelligences. The animal kingdom is extremely diverse in the brains that exist, and there's lots of different niches of nature and some animals have overdeveloped visual cortex or other kind of parts." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On ai-coding-productivity-paradox:
"I do have a cautiously optimistic view of this in software engineering, where it does seem to me like the demand for software will be extremely large and it's just become a lot cheaper. So I do think that for quite some time, it's very hard to forecast. But it does seem to me like right now, at least locally, there's going to be more demand for software because software is amazing... if the barrier comes down, then actually you have the Jevons paradox, which is like, you know, actually the demand for software actually goes up, up. It's cheaper and there's more powerful. The classical example of this always is the ATMs and the bank tellers, because there was a lot of fear that ATMs and computers basically would displace tellers. But what happened is they made the cost of operation of a bank branch much cheaper as there were more bank branches, so there were more tellers." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On autoresearch-recursive-self-improvement:
"I did spend a bunch of time going around opening eye and I was like, you guys realize if we're successful. Like we're all out of job. Like we're just building automation for Sam or something like that. Or the board, I'm not sure, but they're just building this automation for the board or the CEO or something like that. And we're all out of our job and maybe contributing on the sides." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On ai-vampire-pattern, ai-macro-signals-2026, future-of-frontend-engineering:
"if the barrier comes down, then actually you have the Jevons paradox, which is like, you know, actually the demand for software actually goes up, up. It's cheaper and there's more powerful. The classical example of this always is the ATMs and the bank tellers, because there was a lot of fear that ATMs and computers basically would displace tellers. But what happened is they made the cost of operation of a bank branch much cheaper as there were more bank branches, so there were more tellers is like the canonical example people cite. But basically it's just Jevons paradox. Something becomes cheaper, so there's a lot of unlocked demand for it." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On future-of-frontend-engineering, ai-macro-signals-2026:
"these jobs are bundles of tasks and some of these tasks can go a lot faster. And so people should think of it as primarily a tool that it is right now. And I think the long term future of that is uncertain. Yeah, it's kind of really hard to forecast, to be honest. And I'm not professionally doing that really. And I think it's the job of economists to do properly... just being, you know, just trying to keep up with it is like the first thing." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On future-of-frontend-engineering, ai-macro-signals-2026:
"Yeah, what is it limited by? Just, I think, everything. Like so many things, even if they don't work, I think to a large extent you feel like it's a skill issue. It's not that the capability is not there, it's that you just haven't found a way to string it together of what's available. Like, I just don't. I didn't give good enough Instructions in the AgentsMD file or whatever it may be." — 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents (2026-03-20)
-
On (no topic linked):
"[Karpathy joining Anthropic to lead a new pre-training team focused on recursive self-improvement]: he led the self driving team at Tesla also... he was probably the first person that really commercialized the Richard Sutton bitter lesson essay when he was leading FSD at Tesla, which was really about the brute force computation... he's been at the wave upon wave of AI... he's one of these really curious people that can be sent off and they'll just go and invent new things. And I think this idea of recursive self learning puts these models on a combination of overdrive and autopilot. — chamath-palihapitiya in 2026-05-22-podcast-all-in-podcast-spacex-s-2t-case-nvidia-s-shock-selloff-america" — 2026-05-22-podcast-all-in-podcast-spacex-s-2t-case-nvidia-s-shock-selloff-america (2026-05-22)
Sources
- 2025-10-17-dwarkesh-patel-andrej-karpathy-summoning-ghosts
- 2026-03-20-no-priors-andrej-karpathy-skill-issue-code-agents
- 2026-05-22-podcast-all-in-podcast-spacex-s-2t-case-nvidia-s-shock-selloff-america
Related
Cross-links — fill in.