How Autoresearch is changing ML research — Zhengyao Jiang, Weco

Autonomous research agents are transforming machine learning engineering by shifting the focus from manual implementation to high-level system design. Aiden, an autonomous agent developed by Weco, demonstrated this potential by securing seven leaderboard records in OpenAI’s Parameter Golf competition, outperforming human participants while using only 4% of the total compute. Rather than replacing human researchers, agents like Aiden excel at executing complex ideas, navigating vast search spaces, and combining disparate research findings into high-performance solutions. This evolution necessitates a shift in human expertise toward crafting robust evaluation metrics and code-based abstractions, which function as the loss functions and architectures for these autonomous systems. As execution becomes commoditized, the value of human contribution lies in setting the creative vision and structural constraints that guide these agents toward meaningful breakthroughs.