RRSI: Regularized Recursive Self-Improvement of Agent Harnesses

Google Research

Research official 3 src. ~1 min

Harness self-improvement (auto-evolving prompts, control flow, tooling around a frozen model) routinely overfits the training tasks. RRSI adds regularization: a temporally annealed edit budget on the proposer, and a selector with a critic that filters benchmark-specific proposals and a pruner that drops low-value changes.

Why it matters

Up to +4.7 points on out-of-distribution benchmarks and 30% fewer policy tokens, showing self-improving agent harnesses can generalize rather than just memorize the evolution split.

Importance: 3/5

Notable paper from Google Research with open code

agents reasoning rl generalization paper

Sources

official RRSI: Regularized Recursive Self-Improvement of Agent Harnesses

secondary HuggingFace Daily Papers — RRSI

official RRSI code (google-research)