Plan-and-execute agents separate high-level planning from step-by-step execution for complex tasks. Architecture: Planner generates task decomposition and execution order, Executor handles individual steps, Replanner adjusts plan based on execution results. Why separate?: Planning requires global reasoning, execution needs local focus, separation enables specialization, easier to debug and modify. Planning phase: Break task into subtasks, identify dependencies, sequence execution, allocate resources/tools. Execution phase: Execute each step, observe results, report completion status, handle errors. Replanning triggers: Step failure, unexpected results, new information discovered, plan completion. Frameworks: LangChain Plan-and-Execute, BabyAGI, AutoGPT variants. Example: "Research topic and write report" → Plan: [search web, gather sources, outline, draft sections, edit] → Execute each → Replan if sources insufficient. Advantages: Better for complex multi-step tasks, more predictable behavior, easier oversight. Trade-offs: Planning overhead for simple tasks, may over-plan, requires good task decomposition ability.