AIDE (AI-Driven Exploration) by Weco AI is a tree-search agent that drafts, debugs, and benchmarks code until a user-defined metric is optimized. It's the same propose-test-decide pattern as AutoResearch, but applied to Kaggle competitions.
AIDE achieved about x the medal rate of the next best autonomous agent across Kaggle competitions on OpenAI's MLE-Bench. On average, it outperformed half of human contestants. Weco AI reports that most tasks cost under \14$ Turbo.
Where AutoResearch searches program space for ML training improvements, AIDE searches program space for competition solutions. Same pattern, different domain.