@0xCodila - Karpathy Workshop Teaches Claude

Open original

@0xCodila on X A promotional X post pitches two free, short workshops on using Anthropic's Claude: a 14-minute session attributed to Andrej Karpathy on using Claude well, and a 25-minute live build of an AI agent from scratch. The post quotes an earlier claim that a team built in three days, with 44 lines of Claude code, what other frameworks took three years to do.

  • The post credits Andrej Karpathy with a 14-minute workshop on how he uses Claude, framed as a shortcut past a 10,000-hour learning curve.
  • The second claim is that 44 lines of Claude code, released open-source, can produce a working agent in three days, versus three years for other frameworks.
  • The post's stated core recipe is models, tools, and a prompt, which it calls all you need to build a full AI agent.

The post is a promotional tweet bundle, and the claims in it are thin. It links to a tweet from @0xcodila that quotes a second claim: an "Amazon Quant at Anthropic" supposedly said "We built in 3 days what took other frameworks 3 years — 44 lines of Claude code," and that the code was open-sourced. The stated premise is that a working AI agent needs only three things: models, tools, and a prompt. There is also a 25-minute live workshop in which the speaker builds an agent from scratch on stage. Attached is a 14-minute "workshop" billed as coming from Andrej Karpathy, with the line "you literally have to put in 10,000 hours to learn Claude." None of the speakers are named beyond those attributions, and the post offers no repository link, no code, and no transcript in the text itself. Treat the attributions with suspicion. Karpathy is a real and respected figure, but a bookmark-and-watch thread that pins a quote on him and uses a vague "one of the smartest quants in AI" label for someone else is the sort of thing that gets reshared with the names doing the heavy lifting. Check the original sources before repeating the 44-line or 3-day figures, since both are the kind of number that sounds great and is hard to verify. The underlying idea, that an agent is essentially a model in a loop with tools and a good prompt, is a fair description of how many agent systems work, so the core point isn't crazy. The specific numbers are what need proof. If you want to actually use this, skip the thread and look for the open-source repo it mentions. A real 44-line agent is quick to read, and you can judge the "models plus tools plus prompt" claim yourself in ten minutes. Save the video links as a lead, not as a source.

Ask about this article