DeepMind alumnus frames Inherent as a self-improving lab for scientific AI

A lab built around recursive improvement

Louis Kirsch, who worked on AI Scientists at DeepMind, has co-founded Inherent. In his May 28 X post, he described the company as a “recursively self-improving lab for scientific AI”. The phrase is doing real work: Inherent is not pitching a generic assistant, but a research institution designed around agents that help generate, test, and improve scientific work.

The post drew more than 57,000 views and links to Inherent’s public site. Separate launch materials describe Inherent as a London-based Public Benefit Corporation and report a $50 million seed round led by Index Ventures and Radical Ventures. The investor list also includes NVentures, Ex/Ante, Metaplanet, Macroscopic, and Mythos Ventures, which gives the launch more weight than a standard stealth-company reveal.

What to watch

The technical question is whether AI agents can improve the research process itself. Coding agents already show value when tasks are clearly scoped inside repositories. Scientific discovery is harder: the agent must decide which hypotheses are worth testing, how to evaluate partial progress, and when a result is meaningful rather than merely plausible.

Kirsch’s account is a personal researcher channel, but the source is material because it connects the new company to DeepMind work and points directly to the company site. The next evidence should be papers, datasets, reproducible agent traces, or public demonstrations of research workflows. The original post is available on X.