Google showcased a suite of new AI research tools at I/O 2026, positioning them as catalysts for scientific discovery and health assistance. The flagship offering, Gemini for Science, bundles agentic coding platforms like Empirical Research Assistance (ERA) and Co‑Scientist, both recently detailed in Nature. By automating hypothesis generation and code variation, the suite promises faster, more precise experimentation across domains.

ERA enables scientists to write expert‑level empirical software, accelerating projects from neuroscience to cosmology. Early releases on GitHub demonstrate predictions of hospital admissions for respiratory illnesses and seasonal runoff forecasts for California river basins. Co‑Scientist acts as a multi‑agent collaborator, running idea tournaments to produce and debate hypotheses with clickable citations, targeting challenges such as antimicrobial resistance and liver fibrosis.

In health, Google extended its AI to the Google Health app, rolling it out to all Fitbit users and adding a personal‑health LLM called Symptom AI. A study of 13,917 participants showed the model produced differential diagnoses that clinicians preferred twice as often as conventional reviews. Pilot trials also reported a 15% increase in user preparedness for doctor visits, underscoring tangible clinical impact.