'Agentic Scientist' pipeline co-authored an ALIFE 2026 paper

A development version of the system co-wrote a paper that was accepted at ALIFE 2026.

That claim sits in the abstract of "The Agentic Scientist", a new arXiv preprint from Robert Tjarko Lange, Luca Grillotti, Yutaro Yamada, Rujikorn Charakorn and David Ha. The system is a distributed pipeline that decomposes computational research into "task drafting, idea generation, planning, sandboxed experimentation, result aggregation, baseline and ablation construction, post-hoc analysis, and report writing," with each stage powered by coding agents and what the authors call "agentic self-verification." Two researchers we follow in the Who's Who tracker shared the link.

A meta-controller orchestrates parallel research agents and enforces rubric verification at each stage. The authors argue this layer, not the underlying model, is the thing that actually makes autonomy work: "reliable scientific autonomy depends not only on base model capabilities, but also on the cross-agent orchestration layer."

The paper names one concrete publication outcome, the ALIFE 2026 acceptance in co-scientist mode, and describes the fully autonomous runs in general terms, as "open problem-solving tasks in machine learning" and scientific discovery "across different domains of computational research." The abstract does not say which base model powered the coding agents, how much human editing went into the ALIFE submission, or how many autonomous runs it took to produce a usable result.