Top AI Product

Google DeepMind’s Co-Scientist just made the jump from research demo to a tool working scientists can actually request — landing a Nature paper and an experimental rollout through Gemini for Science. It’s a multi-agent system built on Gemini that generates, debates, ranks, and evolves scientific hypotheses.

## A coalition of agents, not one model

Rather than a single model answering questions, Co-Scientist orchestrates specialised agents that argue against each other. They propose hypotheses, critique and rank them in a tournament-style process, and refine the survivors against scientific literature and more than 30 structured life-science databases. The design treats hypothesis generation as an adversarial, iterative search rather than a one-shot generation.

## Why it matters

The headline isn’t the architecture — it’s the wet-lab validation. Co-Scientist proposed novel drug-repurposing candidates for liver fibrosis that held up in laboratory experiments, and predicted antimicrobial-resistance mechanisms that matched results before they were published. That’s the bar that separates a research assistant from a research accelerant: not summarising what’s known, but generating testable, correct ideas humans hadn’t reached yet. Access opens through Google Labs, with an enterprise path via Google Cloud.