Anthropic engineer on self-improving agents
Anthropic engineer says 90% of their engineers use loops and 'dreaming' to build self-improving agentic systems. Key advice: close the agent loop and give an agent a way to verify its own output. In a 30-minute session, a team member shows how to build a self-improving agent using Claude, loops, dreaming, and CLAUDE.md. Playbook available after the talk. #Anthropic
