Tools Andrej Karpathy has publicly discussed or demonstrated using. Curated by Magpie's editorial team from podcasts, posts, and public stacks.
About
Andrej Karpathy is a founding member of OpenAI, former Senior Director of AI at Tesla, and now founder of Eureka Labs, an AI-native education company. He narrates his stack openly: long YouTube lectures and X threads where he names the exact tools he uses and how.
Karpathy publicly switched from GitHub Copilot to Cursor with Sonnet 3.5, posting that the move was 'a net win' and that 'most of my programming is now writing English'. He coined the term 'vibe coding' for an LLM-driven workflow where 'you fully give in to the vibes, embrace exponentials, and forget that the code even exists', citing Cursor Composer with Sonnet as the tooling that makes it possible. By early 2026 he reported 80% agent coding versus 20% manual edits, having flipped from the inverse ratio just months earlier.
Programming is changing so fast... I'm trying VS Code Cursor + Sonnet 3.5 instead of GitHub Copilot again and I think it's now a net win. Just empirically, over the last few days most of my 'programming' is now writing English (prompting and then reviewing and editing the...)
Karpathy has been a consistent Anthropic Claude advocate, pairing it with Cursor for production coding via Sonnet. He's also shared vibe-coding observations from 'claude coding quite a bit last few weeks' and noted his manual-to-agent coding ratio inverted from 80/20 to 20/80 in roughly two months. He frames Claude as the model layer behind the agent transition rather than a general chat assistant.
Karpathy is a founding member of OpenAI and uses ChatGPT continuously for development and education work. He's published lectures on his YouTube channel reproducing GPT-2 from scratch, walking learners through the architecture his own product is built on. ChatGPT is the canonical example he reaches for when discussing LLM behavior on his channel.
Karpathy used GitHub Copilot as his prior coding-assistant default before switching to Cursor + Sonnet 3.5. He still references it as the baseline for AI-assisted coding when discussing the broader evolution of pair-programming tools, and his framing was honest about Copilot being good enough until Cursor's agent loop pulled ahead.
Karpathy uses Perplexity for research-heavy queries and has named it as part of his daily AI stack on X. He frames it as the cited-source answer engine that fills the gap between general LLMs and traditional search for technical research workflows.
These tools have been mentioned by the people featured in podcasts, interviews, posts, and other public media. Source links and citations are included wherever possible. Magpie AI has no direct relationship with the people on these pages. Spot a missing tool or a wrong attribution? Suggest a correction →