Anthropic Study Finds Data Interfaces, Not Reasoning, Constrain Biology AI Agents

Anthropic research found that the main bottleneck for biology AI agents is not model reasoning but database interfaces that cannot support effective retrieval. The finding matters because improving tools and access to data may do more to enhance performance on scientific tasks than simply scaling up models. The study did not involve any investment amount.

The Anthropic team recently developed gget virus, a tool that redesigns how viral data are queried and retrieved. It raised model accuracy on the VirBench biology benchmark to more than 90%. Available information did not disclose the study's publication date or the amount invested, but the results indicate that optimizing data interfaces can significantly unlock the capabilities of AI agents.