AI in Science: Early Insights (Google DeepMind, September 2026): scientists save nearly 7 hours a week with AI
Google, Google DeepMind and MIT FutureTech combine three data sources: a sample of 15 million Gemini interactions, an inventory of more than 2,600 specialized scientific AI models, and a survey of over 600 working scientists, mapped onto a new taxonomy of research tasks. Scientists use AI more than most other occupations, and nearly half of those surveyed use it every day. General LLMs handle analysis, coding and manuscript drafting, while specialized models supply domain predictions, data generation and classification, so the two act as complements rather than substitutes.
Respondents report saving close to 7 hours a week, and most of that time goes back into more research. The catch is that the bottleneck moves downstream: researchers now report a backlog of untested hypotheses and heavy demand for verifying model output.
For anyone building AI for research or technical work, the opportunity is in the verification, experiment and validation layer, where the queue is forming.