AIDE²: First Evidence of Recursive Self-Improvement | Weco AI

weco.ai · 4,122 words · saved by 2 readers

We ran autoresearch on autoresearch: an outer loop rewriting its own research agent for 100 unattended steps. In eight days it discovered agents that beat our two years of hand-tuning on held-out benchmarks, and taught itself to reward hack less.

AIDE²: The First Evidence of Recursive Self-Improvement July 14, 2026 • By Weco Team Share We built a recursive self-improvement (RSI) system by running autoresearch on autoresearch. The system, AIDE 2 , took eight days to discover a better autoresearch harness than the one we built over the last two years. Fully autonomously, AIDE 2 designed a novel search algorithm, reduced the prompt size by 16×, and built a layered system against reward hacking. AIDE 2 has two autoresearch loops: An inner loop, just like a normal autoresearch agent, optimizing code against an eval. An outer loop, optimizin

AIDE²: First Evidence of Recursive Self-Improvement | Weco AI

saved by

related reading