Plan-and-Act

Plan-and-Act is an AI agent pattern that splits problem-solving into two phases: plan the whole sequence of steps first, then execute them one by one. Unlike ReAct, where thinking and acting interleave on every step, Plan-and-Act commits to a plan upfront and only replans when conditions change or new information appears. It suits multi-step workflows where predictability, resource budgeting, and control matter — long-running automation, robotics, enterprise pipelines.