Sakhanda Wire
NVDA $153.59 -1.64% MSFT $507.00 -2.12% GOOGL $279.98 -1.32% META $619.94 -1.03% AMZN $256.62 -0.44%
← Back to the news

Google Research releases RRSI for self-improving AI agent harnesses

Google Cloud AI Research, together with UNC-Chapel Hill, Stanford and Washington University in St. Louis, has released RRSI (Regularized Recursive Self-Improvement), an Apache 2.0 research framework for adapting an LLM agent’s harness. The system can modify prompts, tools, memory, control flow and sub-agents while keeping the model weights frozen. It requires Python 3.10+ and accepts LiteLLM model strings; its defaults assume Claude Opus 4.8 on Vertex AI.

RRSI is designed to limit overfitting during repeated harness evolution. Its mechanisms include a leakage critic, a noise-adjusted performance floor, evidence-based edit tracking, a cost rule and pruning of components that no longer improve results. An annealed edit budget permits broader changes early and more targeted edits later.

On Terminal-Bench 2.1, the framework raised performance from 74.2% to 80.2%, while SWE-bench Verified, which was not used for selection, increased from 82.0% to 83.8%. The three-benchmark out-of-distribution average reached 43.6, compared with 39.7 for the unchanged harness. In an agentic workspace test, RRSI used 2.42 million policy tokens per trial, versus 3.80 million for unregularized evolution. The abstract reports this as 30% fewer tokens, while the project page says 36%.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news