Planning vs Reacting

Planning

Part of: AI Agents & Tool Use

Two agents get the same job: "research three competitors and write a summary." The first writes a full plan up front, list competitors, research each, then summarize, and marches through it. The second takes one step, looks at the result, decides the next step, and repeats. Both can work. They fail differently, cost differently, and shine on different tasks. This is the central design choice in agent building: plan-then-execute versus reactive looping. What it is Plan-then-execute means the agent first turns the goal into a complete ordered plan, then carries it out step by step. The thinking happens up front, in one big burst. Reacting (the ReAct style from earlier) means the agent decides only the next step, executes it, observes the result, and decides again. The thinking is spread across the whole task, one step at a time. The difference is when the agent commits. A planner commits to the whole route before moving. A reactor commits to one move, sees what happens, then recommits. How it works You can see the contrast in how each handles the same plan when a step fails: The planner is efficient when the world is predictable: it thinks once and executes fast. The reactor is relia

Challenge: Planner vs Reactor