Agents in the Wild

Google's AI co-scientist proposed the same superbug hypothesis an Imperial College team spent years on

Google's multi-agent AI co-scientist was asked how some antibiotic-resistant bacteria spread genes. In about two days it proposed the mechanism that Imperial College researchers had found but not yet published.

Organization
Google Research & Imperial College London
Industry
Life Sciences
When
Feb 2025
Evidence
Research paper
Agents and tools
AI co-scientistGemini 2.0

Reported outcome

The system's top hypothesis matched the Imperial team's unpublished finding about how bacteria spread resistance genes.

  • ~2 days (reported)
    Time to hypothesis
  • 6 specialized roles
    Agents in the system

Caveats

Described in a paper with a stated method. Peer review status varies.

  • This is one test where the answer was already known. It does not show the system finds new truths reliably.
  • Google's authors list limits: literature review, fact-checking, and larger expert evaluations need work.
  • The 'decade vs. two days' framing comes from press coverage, not from a controlled comparison.

What the agent does

A scientist states a research goal. Several agents with different jobs (Generation, Reflection, Ranking, Evolution, Proximity, Meta-review) propose, critique, rank, and improve hypotheses in a loop. The scientist can give feedback at any time.

Why it matters

The best test of a research agent is a question where experts already know the answer, but the answer is not public. This is one of the first public tests of that kind.

What to copy

  • Split "come up with ideas" and "criticize ideas" into separate agents.
  • Test research agents on questions you have already solved.
  • Ask for several ranked hypotheses, not one answer.

Starter prompt

Written by us.

Research goal: {goal}
Known background (with citations): {background}

1. Propose 5 distinct hypotheses. For each: mechanism, a key prediction,
   and the cheapest experiment that could disprove it.
2. Now act as a harsh reviewer. Score each 1-10 for novelty and
   testability. Point out any claim that lacks a citation.
3. Rank them and explain the top choice in 3 sentences.

Sources

  1. Accelerating scientific breakthroughs with an AI co-scientist — Google Research, Feb 2025
  2. Google's AI co-scientist could enhance research, say Imperial researchers — Imperial College London, Feb 2025

Added Oct 16, 2026. Spotted an error? Send a correction.

Get new use cases weekly

One short email. Sources included.

Related use cases