Accessible with the Expo Explorer pass and above.
What happens when your agent decides its existing tools aren't good enough and writes new ones? Self-modifying agents can generate, test, and deploy their own tool implementations at runtime, adapting to problems they weren't explicitly programmed to solve. In this session, we'll demo a live agent that forges its own tools on the fly, discuss the safety boundaries you need, and explore where this pattern makes sense (and where it absolutely doesn't).