David Ha's third-generation research system assigns coding agents parallel experiments, baselines and ablations, while researchers remain responsible for judging the results.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [X](https://x.com/hardmaru/status/2110739454948520260)
Why it matters #
Sakana is treating evidence checks and human review as core components of an AI research agent. Its early results show what the workflow can automate, while the mis-scored candidate exposes why the scientific judgment still cannot be handed off.
Sakana AI has released The Agentic Scientist, a third-generation system that coordinates coding agents across research tasks and adds automated baselines and ablations to test whether an apparent result holds up. Co-founder and CEO David Ha (@hardmaru) announced the release on October 15th; the paper had been submitted to arXiv two days earlier, on October 13th. The code is available under an Apache 2.0 license.…