Boris Cherny: We Cut 80% of Claude Code’s Prompt

Y Combinator

Summary

Boris Cherny, creator of Claude Code, discusses the paradigm shift in building AI-agentic products with the release of Claude 3.5 Opus. He emphasizes that modern, more intelligent models require less complex system prompts, often benefiting from the removal of 80% of legacy instructions. Cherny advocates for an empirical, scientific approach where developers act as 'product overhang' solvers, treating AI as a living creature that needs to be tested through constant ablations. He highlights the power of dynamic workflows, test-time compute, and autonomous maintenance routines that allow agents to handle complex coding tasks, self-heal, and innovate over extended periods.

Chapters

  1. 0:07Introduction to Claude Code
  2. 0:21Model Performance and Opus 5
  3. 2:45Solving Prompt Injection
  4. 3:30Deleting the System Prompt
  5. 5:55Ablation as a Development Strategy
  6. 7:39Rebuilding System Prompts
  7. 10:47The Concept of Product Overhang
  8. 14:52Challenging Tasks for Models
  9. 17:39Dynamic Workflows in Practice
  10. 20:13Verification and Prompt Engineering
  11. 28:12Building Agentic Maintenance
  12. 30:25The Future of Coding Skills
  13. 32:39Practical Education
  14. 34:54Final Gift Announcement

Topics covered

  • Performance of Opus 5
  • Prompt Injection Defense

+6 more topics (full content when you follow the channel)

Transcript excerpt

[music] >> All right, Boris. We're so excited to have you here, the creator of Claude code. >> Thank you. >> [applause and cheering] >> It's great to be here. >> Fresh off the press. You guys just shipped Opus 5 yesterday. >> Yes. >> [applause and cheering] >> And it seems that model performance keeps accelerating. You guys got and took Arc AGI 3 to 30%, which is incredible. >> Yes. >> And for context, before the best score was in in the low single digits or low teens, right? What can Opus 5 do now that it couldn't versus the previous version? >> Yeah, there's um there's a lot that goes into every new model and there's a lot of new capabilities that we teach and uh get the model to do. Whenever you do model training, you try to teach a whole bunch of different things and most often it doesn't work. But some subset of the things the model does learn and sometimes it also surprises you. It…