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.

Every tip in Andrej Karpathy's post is gold. I use the HTML one a lot. When I'm learning something new or trying to understand complex code, I ask the agent to build me a visual HTML page. Here I was learning SEO and GEO. The agent turned the key... show more

Kai

14,277 subscribers

51,534 views • 7 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

17 Comments

CODIFY7 days ago

@karpathy Thanks for sharing this!

Esen Espinosa7 days ago

@karpathy I literally ask AI to make me complete career paths for me to tackle. It took a couple weeks, but is possible and easy to read.

Lochin7 days ago

@karpathy How about the cost? When you ask to generate something in HTML, it burns a lot of token or do you use alternative free AI? Because i try to not buy a pro plan and paid subscription, can i still manage to build it in HTML?

Rise of AI6 days ago

@karpathy A nice next step: ask the page to make you predict the result before revealing it. Clicking through an explanation can feel like understanding; changing an input, making a prediction and getting it wrong shows you exactly what you haven’t learned yet.

@hemanth6 days ago

@karpathy Yeah, trying to create an HTML communication skill, but the thing is, I'm not able to create a very good skill that can convey information beautifully. Any tips?

Asmir7 days ago

@karpathy yh the clickable page is nice but i'd be careful, if you can't spot a wrong claim yet then a polished page makes it feel checked. seo and geo advice is half guesswork so i'd make it put a source next to every claim

Kai7 days ago

@karpathy Yeah, hallucinations are definitely a concern here.

Hugo Kemnitz7 days ago

Ask your Agent what he thinks about it in comparison to md files. It is BS. Token consumption will go up due to the waste in the html files. Limited word catalog of the aerospace standard will hurt your brain if you’re semi intelligent. And your agents will struggle to find content specific word.

Rohan7 days ago

@karpathy Asking for an HTML page forces the agent to commit to a structure, so gaps in its understanding show up as missing boxes. Did the SEO and GEO page get anything wrong that you only noticed because it was drawn out?

sofie j,6 days ago

@karpathy this is how i end up with 40 tabs open and a new hobby every week

Yurii Filipchuk🇺🇦 — e/acc6 days ago

@karpathy custom textbook for an audience of one.

Danish Khan7 days ago

@karpathy That airlines manual thingy drastically changed outputs for me, even on smaller models.

TheAgenticHQ6 days ago

@karpathy same, the html pages click better than walls of text. did this for a messy api last week and it finally stuck. your seo one looks clean too.

Directionally Correct6 days ago

@karpathy Yeah, I will literally never make a PowerPoint again 😂

Jordan Lee6 days ago

@karpathy yeah I’d take a clickable page over 50 pages of docs any day lol

Darren Li6 days ago

@karpathy Same here. Asking the agent for an interactive HTML page instead of a wall of text changed how fast I pick up new stuff.

Florian Hack7 days ago

@karpathy Fanboy Post and Late to the Party

Related Videos

My favorite way of interacting with Claude Code is to have it generate static HTML files as outputs (reports, explorations, code structure, mockups etc.) I wanted to iterate on the file by commenting in browser and having Claude update the output live. So, I built this Claude Skill👇 How it works: - Install Claude Code skill (ask it to clone repo) - Build an HTML page for anything (e.g. research coding agents and generate HTML report) - Ask it to make the page interactive That's it. CC will launch a localhost server and allow you to then leave comments on the page itself and once it updates, will give you a tour of changes. It's like Google Docs kind of comments/iteration but for HTML pages.

My favorite way of interacting with Claude Code is to have it generate static HTML files as outputs (reports, explorations, code structure, mockups etc.) I wanted to iterate on the file by commenting in browser and having Claude update the output live. So, I built this Claude Skill👇 How it works: - Install Claude Code skill (ask it to clone repo) - Build an HTML page for anything (e.g. research coding agents and generate HTML report) - Ask it to make the page interactive That's it. CC will launch a localhost server and allow you to then leave comments on the page itself and once it updates, will give you a tour of changes. It's like Google Docs kind of comments/iteration but for HTML pages.

Paras Chopra

216,409 views • 4 months ago

my new favorite workflow: having AI generate clickable briefs for example I had it research every piece of metadata I could use for a design and spin it up as an HTML page so instead of a wall of text I can "see" each concept but the cool part is I can just scroll through and select anything I want to use Claude then opens the browser to read my selections and pulls that context into the chat so I can just say "ok explore 5 variations based on the metadata I selected" and I'm off to the races 🚀

my new favorite workflow: having AI generate clickable briefs for example I had it research every piece of metadata I could use for a design and spin it up as an HTML page so instead of a wall of text I can "see" each concept but the cool part is I can just scroll through and select anything I want to use Claude then opens the browser to read my selections and pulls that context into the chat so I can just say "ok explore 5 variations based on the metadata I selected" and I'm off to the races 🚀

Ridd 🤿

25,387 views • 3 months ago

Quick video on how I use Linear Agent in product work. For feature requests, I want to understand the broader pattern, not just react to one ask. Here, it pulled from 40k+ customer requests to help me think through whether Linear should have team docs.

Quick video on how I use Linear Agent in product work. For feature requests, I want to understand the broader pattern, not just react to one ask. Here, it pulled from 40k+ customer requests to help me think through whether Linear should have team docs.

Karri Saarinen

238,918 views • 6 months ago

Why HTML turned out to be the foundation for agentic video making from Bin Liu: “We’ve been trying to build a video agent. However, we learned the hard way that agents have no visual intelligence. So that’s when we turned to code. HTML is the LLM’s native language. LLMs can express not only information, but also visual aesthetics through HTML, CSS, and JavaScript. Footage, images, assets, SVGs can all sit on top of HTML." 📌 As an example, see how you can make a product video from any website URL here:

Why HTML turned out to be the foundation for agentic video making from Bin Liu: “We’ve been trying to build a video agent. However, we learned the hard way that agents have no visual intelligence. So that’s when we turned to code. HTML is the LLM’s native language. LLMs can express not only information, but also visual aesthetics through HTML, CSS, and JavaScript. Footage, images, assets, SVGs can all sit on top of HTML." 📌 As an example, see how you can make a product video from any website URL here:

Peter Yang

16,653 views • 3 months ago

When I heard about the Copilot SDK, my first thought was whether or not I could build an agent to control my desktop. So I did that. Being able to build and pass custom tools to the agent is op.

When I heard about the Copilot SDK, my first thought was whether or not I could build an agent to control my desktop. So I did that. Being able to build and pass custom tools to the agent is op.

Burke Holland

82,970 views • 8 months ago

receiving a lot of good words about lavish-axi (ty!) in v0.1.27 i just made it installable and invocable as a skill "/lavish" or with a prompt "/lavish let's discuss what options we have" this is the quickest way to upgrade your workflow from reading long wall of text and markdown to interactive html artifacts which are a lot more pleasant and efficient to work with why not just ask the agent to write an html? you'll know when you need to give some feedback to the agent about things in the html and realize there's no good way to make those edits and iterations. lavish gives you the ability to work with the agent via the html itself if you have been using lavish already, make sure to upgrade to latest version! quite a few good bug fixes shipped recently that would help the agent generate correct styling and poll more reliably

receiving a lot of good words about lavish-axi (ty!) in v0.1.27 i just made it installable and invocable as a skill "/lavish" or with a prompt "/lavish let's discuss what options we have" this is the quickest way to upgrade your workflow from reading long wall of text and markdown to interactive html artifacts which are a lot more pleasant and efficient to work with why not just ask the agent to write an html? you'll know when you need to give some feedback to the agent about things in the html and realize there's no good way to make those edits and iterations. lavish gives you the ability to work with the agent via the html itself if you have been using lavish already, make sure to upgrade to latest version! quite a few good bug fixes shipped recently that would help the agent generate correct styling and poll more reliably

Kun Chen

14,584 views • 4 months ago

here's me using design mode in cursor for 10 minutes to build out a new page on our website the starting point is from a one-shot cloud agent that i pulled down locally to iterate. after polish, i send it up to cloud again and let another agent QA it for me so its prod ready

here's me using design mode in cursor for 10 minutes to build out a new page on our website the starting point is from a one-shot cloud agent that i pulled down locally to iterate. after polish, i send it up to cloud again and let another agent QA it for me so its prod ready

eric zakariasson

100,449 views • 5 months ago

Decided that I was going to build an app today using AI. I have no formal CS training and I’ve never coded before. In a just a couple hours, I built Tetris on iOS using ChapGPT. Awesome experience… 1) I asked Chat GPT to teach me along the way (vs just write all the code) 2) when I screwed up, I could send it screenshots and it would tell me what I did wrong 3) It made something intimidating - learning a new language - super approachable 4) ultimately great results: The game works! As we roll into 2025, I’d encourage you to try to learn something alongside AI. More than just the task, you’ll discover that we have some insanely powerful tools now at our finger tips

Decided that I was going to build an app today using AI. I have no formal CS training and I’ve never coded before. In a just a couple hours, I built Tetris on iOS using ChapGPT. Awesome experience… 1) I asked Chat GPT to teach me along the way (vs just write all the code) 2) when I screwed up, I could send it screenshots and it would tell me what I did wrong 3) It made something intimidating - learning a new language - super approachable 4) ultimately great results: The game works! As we roll into 2025, I’d encourage you to try to learn something alongside AI. More than just the task, you’ll discover that we have some insanely powerful tools now at our finger tips

Will Ahmed

63,473 views • 1 year ago

Every time I found myself in an unfamiliar situation in my cricketing career, I never let it get me down. Instead, I adapted to it by learning. Early in my career, the T20 format was introduced. I trained hard for it and learnt to play it. Learning made me confident in something I had never done. It is this very habit of learning that made me confident enough to play all the formats in different conditions around the world. The learning didn’t stop even after I retired. I learnt to be a coach and a mentor. That’s why I like what I have done with the people at CoinDCX : India Ka Crypto Coach. We share a common belief “Learning is the only way to succeed.” #ConfidentCryptoStart #Ad

Every time I found myself in an unfamiliar situation in my cricketing career, I never let it get me down. Instead, I adapted to it by learning. Early in my career, the T20 format was introduced. I trained hard for it and learnt to play it. Learning made me confident in something I had never done. It is this very habit of learning that made me confident enough to play all the formats in different conditions around the world. The learning didn’t stop even after I retired. I learnt to be a coach and a mentor. That’s why I like what I have done with the people at CoinDCX : India Ka Crypto Coach. We share a common belief “Learning is the only way to succeed.” #ConfidentCryptoStart #Ad

Gautam Gambhir

330,891 views • 1 year ago

ROSALÍA talks about what it was like to learn 13 languages to sing the songs on her new album “LUX”. “I speak a little Catalan because it’s my mother tongue. I speak a little Spanish and barely any English, but, you know… I love learning. I love learning things. So I wanted to learn different languages for a long time. I’ve been practicing on Duolingo, you know. But sometimes I’m like, ‘I need to do more.’ So while I was trying to figure out what the concept would be, because I was curious about learning from other cultures and how spirituality can be understood in many different places around the world in different ways, I was learning about different saints from all over the world. And I was reading, reading, reading so many hagiographies, hagiographies. Then I thought, ‘OK, let me put a map on the wall.’ Let me pin different stories, different women from around the world. And if I was making a song inspired by Joan of Arc, then I would do it in French. So depending on the story I wanted to reference, I would use one language or another. And then I would write about 20 different verse options and send them to a translator, a proper translator, get them back, tweak things, record, and then send them to people who would correct the pronunciation, all of it.”

ROSALÍA talks about what it was like to learn 13 languages to sing the songs on her new album “LUX”. “I speak a little Catalan because it’s my mother tongue. I speak a little Spanish and barely any English, but, you know… I love learning. I love learning things. So I wanted to learn different languages for a long time. I’ve been practicing on Duolingo, you know. But sometimes I’m like, ‘I need to do more.’ So while I was trying to figure out what the concept would be, because I was curious about learning from other cultures and how spirituality can be understood in many different places around the world in different ways, I was learning about different saints from all over the world. And I was reading, reading, reading so many hagiographies, hagiographies. Then I thought, ‘OK, let me put a map on the wall.’ Let me pin different stories, different women from around the world. And if I was making a song inspired by Joan of Arc, then I would do it in French. So depending on the story I wanted to reference, I would use one language or another. And then I would write about 20 different verse options and send them to a translator, a proper translator, get them back, tweak things, record, and then send them to people who would correct the pronunciation, all of it.”

ROSALÍA UPDATE

1,821,484 views • 10 months ago

I feel like every time I looked up in 2025 I was reminded of my “why?” My “why” is to try to tell the truth in a way that lets you recognize your own story. It is to honor where I came from while still trying to build something that is bigger than me. It is to use my voice in a way that bridges the gap between the hurt and the healing. It is to take the meaning out of the Whirlwind instead of being swallowed by it. Music ain’t just something I do. It’s how I try to make sense of the world… and since I have yet to make complete sense of the world, I guess I’ll be doing this til the day I can’t. 🩵 Here is to another year of learning that momentum is a gift and not guaranteed. That when dreams move faster than you do, stay rooted and let the Lord carry you. Here is the bird’s eye view of 2025. 🐤 To sum it up, we bent it over our knee and wore its little a$$ out.

I feel like every time I looked up in 2025 I was reminded of my “why?” My “why” is to try to tell the truth in a way that lets you recognize your own story. It is to honor where I came from while still trying to build something that is bigger than me. It is to use my voice in a way that bridges the gap between the hurt and the healing. It is to take the meaning out of the Whirlwind instead of being swallowed by it. Music ain’t just something I do. It’s how I try to make sense of the world… and since I have yet to make complete sense of the world, I guess I’ll be doing this til the day I can’t. 🩵 Here is to another year of learning that momentum is a gift and not guaranteed. That when dreams move faster than you do, stay rooted and let the Lord carry you. Here is the bird’s eye view of 2025. 🐤 To sum it up, we bent it over our knee and wore its little a$$ out.

Lainey Wilson

63,205 views • 9 months ago

NS Rb: Thank you for remembering that it's been six years. Right now, this feels like the best day of my life. Starting all the way back from the VALORANT beta in 2020, I think I always viewed myself more as someone who was still learning. Honestly, I never really thought of myself as someone who could just guarantee that I was a great player. So from the members I played with starting in Vision Strikers through DRX up until 2023, I think I spent those three years learning. Then through that learning, when I went to TEC in VCT CN, I thought a lot about how to use what I had learned, not just in-game but also more broadly in life. And that eventually brought me here to Nongshim and to this moment. In the end, though, the result came down to learning, studying, teammates covering for one another, and most importantly, as obvious as it sounds, hard work. (via EndPoem)

NS Rb: Thank you for remembering that it's been six years. Right now, this feels like the best day of my life. Starting all the way back from the VALORANT beta in 2020, I think I always viewed myself more as someone who was still learning. Honestly, I never really thought of myself as someone who could just guarantee that I was a great player. So from the members I played with starting in Vision Strikers through DRX up until 2023, I think I spent those three years learning. Then through that learning, when I went to TEC in VCT CN, I thought a lot about how to use what I had learned, not just in-game but also more broadly in life. And that eventually brought me here to Nongshim and to this moment. In the end, though, the result came down to learning, studying, teammates covering for one another, and most importantly, as obvious as it sounds, hard work. (via EndPoem)

VALO2ASIA

58,285 views • 7 months ago

Introducing: Tesla CLI/Claude Code Skill/OpenClaw and Hermes skill from the Printing Press. - "Unlock the car" and "turn on dog mode" as one-line commands, callable from your phone or laptop - Agent: "during winter school days, defrost my car at 7:50 every weekday before school dropoff" - Charging cost ledger - Supercharger queue watcher pageable from an agent - Your signing key stays on YOUR host - Much more Fun fact: when I got my first ClawdBot, Tesla was one of the first skills I made. But I could only get it to work with my older Tesla. Now that I have the Printing Press, I was able to build what I wanted soup to nuts and now it's here.

Introducing: Tesla CLI/Claude Code Skill/OpenClaw and Hermes skill from the Printing Press. - "Unlock the car" and "turn on dog mode" as one-line commands, callable from your phone or laptop - Agent: "during winter school days, defrost my car at 7:50 every weekday before school dropoff" - Charging cost ledger - Supercharger queue watcher pageable from an agent - Your signing key stays on YOUR host - Much more Fun fact: when I got my first ClawdBot, Tesla was one of the first skills I made. But I could only get it to work with my older Tesla. Now that I have the Printing Press, I was able to build what I wanted soup to nuts and now it's here.

Matt Van Horn

129,146 views • 4 months ago

This was the whole product, but I didn't have the patience to finish it 😅. I'm an animation student, and this was my first ''big'' personal project, so I got really caught up while learning along the way, and believe me, the toon boom file is a mess. But I learned a lot!

This was the whole product, but I didn't have the patience to finish it 😅. I'm an animation student, and this was my first ''big'' personal project, so I got really caught up while learning along the way, and believe me, the toon boom file is a mess. But I learned a lot!

Nonie2.0

914,972 views • 3 years ago

I gave Hermes OS one voice command and watched it build a website “Create a website for a hair salon.” My agent started writing code, moving through windows, and building the site right in front of me. I could follow the whole process in real time I just added native voice commands to Hermes OS, and there’s an important difference: Controlling the OS uses zero model tokens. Those commands run natively. You can control your entire computer through your voice Tokens only come into play when I ask Hermes Agent to do a task This is what I wanted Hermes OS to feel like Tell my computer what I need. Watch it get to work.

I gave Hermes OS one voice command and watched it build a website “Create a website for a hair salon.” My agent started writing code, moving through windows, and building the site right in front of me. I could follow the whole process in real time I just added native voice commands to Hermes OS, and there’s an important difference: Controlling the OS uses zero model tokens. Those commands run natively. You can control your entire computer through your voice Tokens only come into play when I ask Hermes Agent to do a task This is what I wanted Hermes OS to feel like Tell my computer what I need. Watch it get to work.

Luke The Dev

20,122 views • 17 days ago

"My kids do not go to school, and have never been to school." "Not because I don't value learning, I value learning very much. I just disagree with the belief that school is a necessary component for learning." "I believe children are always learning, and when allowed to pursue something they are actually interested in, something they have a need to learn, something that is relevant to their lives and has real world context, I believe that learning is exponentially more effective than the coercive curriculums of the conventional educational model." Agree or disagree? 🤔

"My kids do not go to school, and have never been to school." "Not because I don't value learning, I value learning very much. I just disagree with the belief that school is a necessary component for learning." "I believe children are always learning, and when allowed to pursue something they are actually interested in, something they have a need to learn, something that is relevant to their lives and has real world context, I believe that learning is exponentially more effective than the coercive curriculums of the conventional educational model." Agree or disagree? 🤔

Wide Awake Media

40,908 views • 1 year ago

For those interested, I also have my own approach to the problem, though I forgot to properly advertise it: While Marc-André Moreau's solution rely on the API, in my case I give direct access to WinDbg to the agent. It allows for instance to have an agent join you in a session to help you figuring something out. Or the agent can still manage the WinDbg instance autonomously from start to finish. I don't know if one solution is better than the other, it probably depends on the use-case. But the more the merrier 😀

For those interested, I also have my own approach to the problem, though I forgot to properly advertise it: While Marc-André Moreau's solution rely on the API, in my case I give direct access to WinDbg to the agent. It allows for instance to have an agent join you in a session to help you figuring something out. Or the agent can still manage the WinDbg instance autonomously from start to finish. I don't know if one solution is better than the other, it probably depends on the use-case. But the more the merrier 😀

Kevin Gosse

12,850 views • 1 month ago

Trying new things and learning new skills brings an overwhelming happiness and excitement. The joy I felt when I tried something new was gave me confidence, energy, and a new perspective ❤️ #mrx #shootingdiaries

Trying new things and learning new skills brings an overwhelming happiness and excitement. The joy I felt when I tried something new was gave me confidence, energy, and a new perspective ❤️ #mrx #shootingdiaries

Athulyaa Ravi

13,535 views • 4 months ago

Three skills I use every day in Claude Code and Codex to solve my hardest problems: 1️⃣ /agent-watchdog When I have one agent like Codex working on a task and I don't fully trust it's going to do everything right, I'll open up another one like Claude Code and tell it to watchdog the Codex thread. You can copy the Codex deep link into Claude Code and it'll look at the prompt you sent, watch the Codex thread until it's done, then compare the Codex solution to how it was planning to solve it and automatically fix anything that Codex missed. It can also test the work of the other agent end-to-end. Similar to the idea of OpenRouter's new Fusion feature, I've definitely found that two models thinking through a problem and checking each other's work can be wildly more impactful than just one. 2️⃣ /plan-arbiter Similar ideas as /agent-watchdog - but with this one you have both make plans, compare plans, negotiate the differences, and make a final plan to execute. I find Claude Code is better at writing plans, but Codex is faster and cheaper to execute on them. Then I usually have Claude Code watchdog the Codex work and fix anything that was missed. 3️⃣ /read-the-damn-docs One thing that drives me crazy with coding agents is they're so reluctant to look up docs. They'll just guess and guess and guess at the right API surface for things, or the right solution to an integration of two things. Once I explicitly tell it to look up the docs, it says "Oh, I see the answer," and it fixes the problem. So I made the /read-the-damn-docs skill. Add it and your agents will know when and how to do efficient web searches to look up docs for the types of problems you really should look up docs for. All of these are totally open source over on my GitHub. If you try them, let me know your feedback. Will link to them below:

Three skills I use every day in Claude Code and Codex to solve my hardest problems: 1️⃣ /agent-watchdog When I have one agent like Codex working on a task and I don't fully trust it's going to do everything right, I'll open up another one like Claude Code and tell it to watchdog the Codex thread. You can copy the Codex deep link into Claude Code and it'll look at the prompt you sent, watch the Codex thread until it's done, then compare the Codex solution to how it was planning to solve it and automatically fix anything that Codex missed. It can also test the work of the other agent end-to-end. Similar to the idea of OpenRouter's new Fusion feature, I've definitely found that two models thinking through a problem and checking each other's work can be wildly more impactful than just one. 2️⃣ /plan-arbiter Similar ideas as /agent-watchdog - but with this one you have both make plans, compare plans, negotiate the differences, and make a final plan to execute. I find Claude Code is better at writing plans, but Codex is faster and cheaper to execute on them. Then I usually have Claude Code watchdog the Codex work and fix anything that was missed. 3️⃣ /read-the-damn-docs One thing that drives me crazy with coding agents is they're so reluctant to look up docs. They'll just guess and guess and guess at the right API surface for things, or the right solution to an integration of two things. Once I explicitly tell it to look up the docs, it says "Oh, I see the answer," and it fixes the problem. So I made the /read-the-damn-docs skill. Add it and your agents will know when and how to do efficient web searches to look up docs for the types of problems you really should look up docs for. All of these are totally open source over on my GitHub. If you try them, let me know your feedback. Will link to them below:

Steve (Builder.io)

43,089 views • 3 months ago

I built a Visual Studio Code extension to turn Visual Studio code into a custom learning environment for my Machine Learning class. Attached, you'll see a 2-minute video of how it works. I haven't tried this with my students yet (the February cohort will be the first to do so), but I think it will be huge. I wish more people would offer a learning experience like this. My next cohort starts in February and you can join at You'll get lifetime access to the best Machine Learning engineering cohort online.

I built a Visual Studio Code extension to turn Visual Studio code into a custom learning environment for my Machine Learning class. Attached, you'll see a 2-minute video of how it works. I haven't tried this with my students yet (the February cohort will be the first to do so), but I think it will be huge. I wish more people would offer a learning experience like this. My next cohort starts in February and you can join at You'll get lifetime access to the best Machine Learning engineering cohort online.

Santiago

82,846 views • 1 year ago