Google shifts emphasis from specialized AI science tools toward agentic systems that conduct research with limited human guidance

At Google I/O, Demis Hassabis, CEO of Google DeepMind, positioned scientific AI as central to the company's vision, highlighting WeatherNext (weather prediction software) that provided advance warning of Hurricane Melissa's landfall in Jamaica. Concurrently, Google announced Gemini for Science, a package uniting LLM-based scientific systems including the hypothesis-generating AI Co-Scientist and algorithm-optimizing AlphaEvolve, which researchers can now apply to access.

The shift reflects a reorientation toward agentic systems—AI that executes research tasks autonomously—rather than specialized tools like AlphaFold (trained for protein structure prediction). A sign of this pivot: John Jumper, Nobel laureate for AlphaFold, is now working on AI coding, which Google notes is key to agentic system success.

Agentic systems are already making independent contributions: this week, OpenAI announced that one of their models disproved an important mathematics conjecture, described by some mathematicians as perhaps the most meaningful contribution generative AI has made to mathematics so far. The model used is a general-purpose reasoning model, not specialized for mathematical research.

Specialized tools remain widely adopted—AlphaFold predictions have been used by over three million researchers worldwide, and Isomorphic Labs (a Google subsidiary developing drugs via AlphaFold technology) raised a $2 billion Series B funding round.