Meet ‘AutoAgent’: The Open-Source Library That Lets an AI Engineer and Optimize Its Own Agent Harness Overnight

Kevin Gu’s thirdlayer.inc released AutoAgent, an open-source system that lets a meta-agent improve its own agent harness.

The system can modify prompts, tools, routing, and orchestration, then keep only changes that boost benchmark scores.

After a 24-hour run, the project reported top results on SpreadsheetBench and TerminalBench.

Its workflow centers on a human-edited directive file, an editable harness file, and an experiment log.