Toward Self-Improving Agentic Systems: Berkeley Summit Talk

furongh · x · 2026-07-31

The author points out that self-improving AI is not just about model capabilities (like better reasoning or learning from outputs); the surrounding infrastructure and deployment system can also be designed for self-improvement.

This includes how an agent allocates compute, evaluates and revises actions, and composes or redesigns workflows for specific tasks. The author will share this vision at the Berkeley RDI Agentic AI Summit in a talk titled "Reasoning as Control: Toward Self-Improving Agentic Systems," covering three levels: Thinking → Action → Workflow.

Original post →

More from coding & agent

coding & agent channel →