Stanford's Paper2Agent turns research papers into AI agents that run real code

Stanford researchers published Paper2Agent in Nature, a system that converts research papers with public GitHub code into AI agents that actually execute the study's code. It succeeded on 74 of 100 bioinformatics papers and hit about 98% accuracy on non-bio tasks.

• 74 of 100 bioRxiv papers converted; ~98% accuracy outside biology
• MIT-licensed, runs via Claude Code or Gemini CLI
• Conversion costs $14–15 per repo and takes 45 minutes to 3 hours
• Requires public GitHub code and runnable tutorial files