Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Andrej Karpathy says when AI agents fail, it's usually a skill issue, not a capability issue You didn't write good enough instructions, didn't set up the right memory tool, or didn't parallelize correctly "the real shift is working in macro actions" One does research, one writes code, one plans,... show more

Haider.

66,931 subscribers

368,897 views • 3 months ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

karpathy on Dwarkesh in Oct. '25: > AI agents don't work > They are not intelligent enough > They are not multi-modal enough > They can't do computer use karpathy on No Priors in March '26: > AI agents fail but it's a skill issue for the humans > You didn't give them the right instructions AI agents absolutely hit an inflection point in December of 2025 and there's no looking back.

karpathy on Dwarkesh in Oct. '25: > AI agents don't work > They are not intelligent enough > They are not multi-modal enough > They can't do computer use karpathy on No Priors in March '26: > AI agents fail but it's a skill issue for the humans > You didn't give them the right instructions AI agents absolutely hit an inflection point in December of 2025 and there's no looking back.

Dan McAteer

136,855 views • 3 months ago

New Andrej Karpathy interview Says AI agent failures stem from user skill, not model capability. Poor instructions cause errors. He suggests delegating 20-minute macro actions like coding and research to parallel agents and reviewing their work. --- "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 together what's available. Like, I didn't give good enough instructions to the agents in the 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 a skill issue when it doesn't work to some extent. You want to see how you can parallelize them, and you want to be a 'Pierce tender,' basically. Pierce famously has a funny photo where he's in front of lots of these Codex agents behind the monitor. They all take about 20 minutes if you run them correctly and use high effort. You have multiple—you know, 10 or 20—pull requests checked out. It's just like you can do much larger macro actions. It's not just, 'Here's a line of code, here's a new function.' It's like, 'Here's a new functionality, delegate it to agent one. Here's a new functionality that's not going to interfere with the other one, give it to agent two.' Then, you try to review their work as best as you can, depending on how much you care about that code. You look for these macro actions that you can manipulate your software repository by. Another agent is doing some research, another agent is writing code, another one is coming up with a plan for some new implementation. Everything just happens in these macro actions over your repository. You're just trying to become really good at it and develop a muscle memory for it. It's very rewarding when it actually works, but it's also a new thing to learn. Hence, the psychosis." --- From No Priors YT channel (link in comment)

New Andrej Karpathy interview Says AI agent failures stem from user skill, not model capability. Poor instructions cause errors. He suggests delegating 20-minute macro actions like coding and research to parallel agents and reviewing their work. --- "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 together what's available. Like, I didn't give good enough instructions to the agents in the 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 a skill issue when it doesn't work to some extent. You want to see how you can parallelize them, and you want to be a 'Pierce tender,' basically. Pierce famously has a funny photo where he's in front of lots of these Codex agents behind the monitor. They all take about 20 minutes if you run them correctly and use high effort. You have multiple—you know, 10 or 20—pull requests checked out. It's just like you can do much larger macro actions. It's not just, 'Here's a line of code, here's a new function.' It's like, 'Here's a new functionality, delegate it to agent one. Here's a new functionality that's not going to interfere with the other one, give it to agent two.' Then, you try to review their work as best as you can, depending on how much you care about that code. You look for these macro actions that you can manipulate your software repository by. Another agent is doing some research, another agent is writing code, another one is coming up with a plan for some new implementation. Everything just happens in these macro actions over your repository. You're just trying to become really good at it and develop a muscle memory for it. It's very rewarding when it actually works, but it's also a new thing to learn. Hence, the psychosis." --- From No Priors YT channel (link in comment)

Rohan Paul

23,090 views • 3 months ago

"The squad is not good enough to survive, the appointment didn't feel like the right one" 💬 Former Leicester striker Alan Smith says he didn't understand why Leicester appointed Ruud van Nistelrooy 🚨

"The squad is not good enough to survive, the appointment didn't feel like the right one" 💬 Former Leicester striker Alan Smith says he didn't understand why Leicester appointed Ruud van Nistelrooy 🚨

Sky Sports News

1,041,273 views • 1 year ago

Is this a skill issue or just bad luck in ARC Raiders? Bro didn't even use his gun properly! 😭

Is this a skill issue or just bad luck in ARC Raiders? Bro didn't even use his gun properly! 😭

ARC Raiders Alerts

61,238 views • 6 months ago

Claude Code is no longer one AI helping you write code. It now creates full agent teams. Multiple AIs. Different roles. Shared tasks. Constant communication. All working at the same time. Like a real dev team… but automated.

Claude Code is no longer one AI helping you write code. It now creates full agent teams. Multiple AIs. Different roles. Shared tasks. Constant communication. All working at the same time. Like a real dev team… but automated.

0xMarioNawfal

62,137 views • 4 months ago

LSP? What is it? Why is it? How is it? No one ever asks how it is, and I didn't in the video either but it's a pretty fun one. Also, we set it up in like 3 lines of code, you'll like that :)

LSP? What is it? Why is it? How is it? No one ever asks how it is, and I didn't in the video either but it's a pretty fun one. Also, we set it up in like 3 lines of code, you'll like that :)

teej dv 🔭

23,553 views • 1 year ago

jeongin said they should ask kickflip kyehoon whether jeongin really gave jisung a kiss or not 😭 🐿: no, he kissed in the hallway and you kissed me in the room 🦊: in my memory, we didn't 🐿: your memory is wrong. the one who does it doesn't know and the one who gets kissed knows

Sensitive content

jeongin said they should ask kickflip kyehoon whether jeongin really gave jisung a kiss or not 😭 🐿: no, he kissed in the hallway and you kissed me in the room 🦊: in my memory, we didn't 🐿: your memory is wrong. the one who does it doesn't know and the one who gets kissed knows

비니

335,184 views • 7 months ago

.Laura Ingraham: "Not one single Republican Senator endorsed [Ken Paxton]. What did you know that the GOP establishment didn't? Me: "It's not good enough to say the right things... What the people want is somebody who is going to do the right thing."

.Laura Ingraham: "Not one single Republican Senator endorsed [Ken Paxton]. What did you know that the GOP establishment didn't? Me: "It's not good enough to say the right things... What the people want is somebody who is going to do the right thing."

Brandon Gill

308,493 views • 1 month ago

claude code now writes its own test plan and runs it on a live iphone. i didn't write the tests. i didn't run the tests. i didn't even open the app. it caught the bug too.

claude code now writes its own test plan and runs it on a live iphone. i didn't write the tests. i didn't run the tests. i didn't even open the app. it caught the bug too.

Landseer Enga

30,714 views • 2 months ago

How a RIB adjustment is done.....especially when you didn't know you had ribs up there, and that it was the issue for you!!!

How a RIB adjustment is done.....especially when you didn't know you had ribs up there, and that it was the issue for you!!!

Capt KEVman!!!

688,495 views • 2 months ago

I built an AI agent that does my weekly stock research while I sleep. It scans NASDAQ, flags movers, writes summaries, and exports slides. I didn't write a single prompt. Here's the framework you're not using yet:

I built an AI agent that does my weekly stock research while I sleep. It scans NASDAQ, flags movers, writes summaries, and exports slides. I didn't write a single prompt. Here's the framework you're not using yet:

Shruti

90,222 views • 4 months ago

Reporter: You say he didn't sign the birthday card, he didn't do the check. Those were documents from the estate. What is the working theory? Leavitt: The president has one of the most famous signatures in the world. The president did not write that letter, did not sign those documents.

Reporter: You say he didn't sign the birthday card, he didn't do the check. Those were documents from the estate. What is the working theory? Leavitt: The president has one of the most famous signatures in the world. The president did not write that letter, did not sign those documents.

Acyn

1,053,690 views • 10 months ago

The cameraman had one job and didn't fail at it

The cameraman had one job and didn't fail at it

The Best Viral Vids🔥

409,868 views • 5 months ago

#WATCH: Jagmeet Singh says our healthcare system is in crisis and he's the only one who is standing up to defend it. Let's recap. Didn't stand up to Trudeau. Didn't stand up for port or rail or post workers. Didn't stand up for Canadians. Stood up for his pension!

#WATCH: Jagmeet Singh says our healthcare system is in crisis and he's the only one who is standing up to defend it. Let's recap. Didn't stand up to Trudeau. Didn't stand up for port or rail or post workers. Didn't stand up for Canadians. Stood up for his pension!

govt.exe is corrupt

38,026 views • 1 year ago

Richard Tice says the number one issue facing the country are small boat arrivals of asylum seekers It's not my number one issue, mine would be the underfunding of the NHS What's your number one issue?

Richard Tice says the number one issue facing the country are small boat arrivals of asylum seekers It's not my number one issue, mine would be the underfunding of the NHS What's your number one issue?

Farrukh

282,879 views • 1 year ago

Rachel didn't put up Kelley b/c she didn't want a spell cast on her, Will says Kelley is the only one he's scared of in the house "who practices witchcraft in 2025?" #bb27

Rachel didn't put up Kelley b/c she didn't want a spell cast on her, Will says Kelley is the only one he's scared of in the house "who practices witchcraft in 2025?" #bb27

#BB27 Feeds Fairy

94,726 views • 10 months ago

Omma AI is wild. You type one sentence and it can create: • a 3D scene • a full website • a working app At the same time. One agent writes code. One builds the 3D model. One designs the interface. That is a very different direction from most AI tools.

Omma AI is wild. You type one sentence and it can create: • a 3D scene • a full website • a working app At the same time. One agent writes code. One builds the 3D model. One designs the interface. That is a very different direction from most AI tools.

Julian Goldie SEO

11,982 views • 3 months ago

Excited to present the LLM-Council skill. Initial idea by Karpathy. I just packaged it as a skill. You can easily spin up a council of LLMs or agents via Fireworks AI. Watch how the new GLM-5 model "deliberates" on other LLMs' thoughts on the big question, "Can LLMs reason?" Things worth paying attention to: New open models like GLM-5 have surprisingly improved on complex reasoning and long-running agentic tasks. The AskUserQuestion tool in Claude Code came in handy to select the council and chairperson. As Andrej Karpathy puts it, it's a really interesting way to get different perspectives from LLMs, which can lead to better decision-making on whatever task you are working on. You can use it for other agentic coding use cases, like evaluation, tool building, designing, and research.

Excited to present the LLM-Council skill. Initial idea by Karpathy. I just packaged it as a skill. You can easily spin up a council of LLMs or agents via Fireworks AI. Watch how the new GLM-5 model "deliberates" on other LLMs' thoughts on the big question, "Can LLMs reason?" Things worth paying attention to: New open models like GLM-5 have surprisingly improved on complex reasoning and long-running agentic tasks. The AskUserQuestion tool in Claude Code came in handy to select the council and chairperson. As Andrej Karpathy puts it, it's a really interesting way to get different perspectives from LLMs, which can lead to better decision-making on whatever task you are working on. You can use it for other agentic coding use cases, like evaluation, tool building, designing, and research.

elvis

39,452 views • 4 months ago