Agentic loops and leaner prompts
This week, we track the rise of continuous, always-on agentic loops, major model releases from Moonshot and Thinking Machines, and a fundamental shift toward leaner prompting strategies.
Signals worth your attention
Weco documents recursive self-improvement in production
AI research lab Weco built AIDE², an autonomous system where one AI agent optimizes another agent's code in a continuous loop without human intervention. Running for 8 days across 100 iterations, the system produced AIDE85, which outperformed a human-engineered agent that Weco's team spent two years building, as reported by AI Edge. This represents one of the first practical, documented cases of recursive self-improvement in production-grade software agents.
Free, unlock with your email. You get the rest of this issue now and every next issue on Friday morning.