GOOGLE HARNESS² FOR RECURSIVE SELF IMPROVEMENT

Summary from the TLDR AI newsletter

Google Research's Harness² lets agents iteratively improve their prompts, tools, memory, and orchestration without changing model weights. It refines the harness for each task and then carries evidence-backed lessons forward to improve future tasks.