Rohan Paul (@rohanpaul_ai)T2·Agents
78 pts
A new Google study finds that self-improving agent harnesses work best when an optimizer model edits one code module at a time under validation and token caps, rather than performing an unchecked full rewrite. On 40 held-out SWE-bench Lite tasks, the gated approach reached 40.0% accuracy at 46.7k tokens per task, versus 42.5% at 451.6k tokens for a full rewrite, nearly 10x the cost. Enforcing rules via prompts instead of code collapsed accuracy to 2.5%.