The system wraps off-the-shelf coding agents such as Codex and Claude Code in a scaffold that evaluates output against predefined rubrics. In autonomous runs, the tool increased Doc-to-LoRA retrieval accuracy from 52% to 98% on multi-needle tasks. Alongside the agent framework, Sakana AI released AutoIllustrate, a pipeline that generates editable vector graphics for scientific figures instead of hallucinated raster images.
Wednesday, October 21, 2026
Story
Sakana AI Releases Agentic Scientist v3 for Automated Scientific Discovery
AI6 days ago
A new orchestration framework enables multiple AI agents to share memory and collaborate on experimental baselines and ablations to accelerate academic discovery. This system, called The Agentic Scientist, is the third iteration of a research tool by Sakana AI and is now open source. Rather than relying on a single agent to write a paper, the framework uses a pool of parallel workers and a meta-controller to manage the research loop from ideation to report writing. An early version of the system contributed to a paper on biological learning rules that was accepted as an oral presentation at ALIFE 2026.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.