StepFun: Deep Research is a Training Problem, Not Scaling

StepFun details how training atomic skills like planning and reflection, rather than simple scaling, enables their AI agent to conduct autonomous research end-to-end.