deepmind open-sources science skills: ai agents meet 30+ databases

google is quietly making a strategic move in the war for scientific talent

deepmind just open-sourced "science skills" – a toolkit that lets ai agents natively talk to 30+ real scientific databases: alphafold, pubmed, clinicaltrials, uniprot, arxiv, ensembl, and more

each "skill" is a structured instruction file that tells an agent exactly how to query a database, parse results, and avoid hallucinating data

• this is google's r&d cost play. instead of throwing more researchers at literature reviews, protein analysis, variant databases, and clinical trial screening – you automate the grunt work. faster research cycles, lower headcount per discovery

• but read the subtext: anthropic and openai are running the same playbook – optimizing their models hard for scientific reasoning, publishing benchmarks on biology and chemistry tasks, sponsoring research tools. all three labs are competing for the same pool of phds and biotech builders who now decide which ai stack to build on

• for founders: if you're building in pharma, biotech, clinical intelligence, or genomics – these skills are free connectors to data pipelines that would take months to build. apache 2.0, use commercially