On Oct 31 Andrej Karpathy published HackerTube (hackertube.org) on his personal blog: a YouTube-style site that only lists the 6,454 YouTube videos recommended in Hacker News comments from 2007 to 2026, grouped into 150 topics. He says Opus 5.5 in Claude Code did the work: it downloaded every HN comment, ran Sonnet over them in a Workflow to narrow things down to about 40K comments recommending a video, wrapped the result in a YouTube UI, then deployed and hardened it on a VPS, all in about 2 hours. His only manual steps were creating a YouTube API key, booting the VPS and buying the domain. His HN submission reached the front page with about 1,005 points and 308 comments.
Key Takeaways
- ✓Source: Karpathy's personal blog post HackerTube (Oct 31), with the site live at hackertube.org
- ✓Scale: from every HN comment it narrowed things down to about 40K comments recommending a YouTube video, giving 6,454 videos across 150 topics from 2007 to 2026 (author-reported)
- ✓Method: Opus 5.5 ran the download and overall pipeline, used Claude Code's Workflow feature to have Sonnet filter comments and drop spam in bulk, then built a YouTube-style front end and deployed and hardened it on a VPS
- ✓Product details: it uses the frame at the 50% mark instead of the uploaded thumbnail, offers account-free For You picks with state kept only in the local browser, supports keyboard shortcuts, and exports the site's data so you can hand it to your own agent to build a variant
- ✓Workflow: Karpathy says Opus 5.5 on Medium effort is his default and he no longer reads or writes code or runs terminal commands himself; the HN post has about 1,005 points and 308 comments (public HN data as of Nov 1, 14:10 SGT)
Key Decision Metrics at a Glance
Heavy Claude Code use: compare subscription limits and API bills
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and the problem
YouTube's recommendations are tuned for clicks, and plenty of technical users complain that exaggerated thumbnails and repetitive topics bury anything good. Andrej Karpathy, an OpenAI founding member and former head of AI at Tesla, wrote on his personal blog on Oct 31 that at 11pm on Friday he opened YouTube and saw a wall of slop, so the next morning he had Opus 5.5 build his own YouTube made only of videos that Hacker News users recommended in comments. The result is hackertube.org. It is Karpathy's personal project, not an official product from Anthropic or any company.
Architecture and how it works
By Karpathy's account, Opus 5.5 first downloaded every HN comment, then used Claude Code's Workflow feature to run Sonnet over them in bulk and strip out spam, ending with about 40K comments that recommend a YouTube video. That maps to 6,454 videos across 150 topics from 2007 to 2026. It then wrapped everything in a YouTube-style UI, and after Karpathy bought a domain and booted a VPS, Claude updated, deployed, secured, hardened and optimized it. A few product choices are deliberate: thumbnails use the frame at the 50% mark instead of clickbait covers; For You picks need no account and keep state only in the local browser session; there are keyboard shortcuts; and the full dataset can be exported, which the author says is so you can give it to your own agent to build your own version.
Benchmarks and measured numbers
This is not a model launch, so there are no benchmarks. The verifiable numbers come from the author and public pages: about 2 hours of work, about 40K comments, 6,454 videos and 150 topics (author-reported). On reach, Karpathy's own HN submission had about 1,005 points and 308 comments as of Nov 1, 14:10 SGT and was one of the day's top stories, and it was widely relayed on X, where swyx noted that Karpathy used Opus 5.5 on Medium effort.
Getting started
The blog's process note matters more to developers than the site itself. Karpathy says he uses the Claude Code desktop app on Mac with Opus 5.5 on Medium effort as his default, and no longer does anything resembling traditional engineering: he doesn't read or write code or run terminal commands, he just talks to Claude. His only manual steps this time were creating an API key in the Google Cloud console to query YouTube metadata, booting a VPS and buying the domain; once the box was up he gave Claude the IP and it handled deployment and hardening. The reusable pattern is a strong model orchestrating while a cheaper model batch-processes the bulk data. To build your own version, export the data from hackertube.org and hand it to your agent, but still check API key permissions and server security yourself.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.