Google Unveils ERA — AI Tool for Scientific Programming, Published in Nature

On May 19, 2026, Google Research announced the publication in the journal Nature of a description of Empirical Research Assistance (ERA), an AI tool for writing and optimizing scientific code at an expert level. ERA uses the Gemini model; it is capable of searching scientific literature, writing code, exploring solutions, combining techniques, and evaluating results using tree search among thousands of options. The system was tested on tasks in genomics, public health, satellite image analysis, neuroscientific forecasting, time series, and mathematics — achieving expert-level performance in all areas. ERA helped create a prototype of Computational Discovery, which is beginning to be deployed under the Gemini for Science initiative; access is opening through a trusted tester program. Additionally, the Hypothesis Generation experiment has been launched based on AI Co-Scientist (described in the same Nature article), along with the Literature Insights tool. In total, eight scientific manuscripts have been prepared using ERA on specific problems, including epidemic forecasting, carbon dioxide monitoring, runoff prediction, solar energy, and macroeconomic retail sales forecasting.