A post argues that problems that are fast and easy to verify represent a small subset of scientific problems but will be most affected by AI. Alongside tackling those, it recommends making verification easier through faster small-molecule synthesis, new assays, new biomarkers, and accelerated clinical trials. For problems that cannot be verified easily, it recommends gathering data. The author adds that many other kinds of work remain important and valuable, but predicts they will benefit less from AI.