Six AI Agent Planning Patterns and Guidance on Choosing Among Them

This developer guide advises starting from the specific task rather than from a popular reasoning pattern when designing AI agents. It outlines options including fixed workflows, plan-then-execute, ReAct-style tool use, routing to specialists, parallel work, and draft-evaluate-revise. The article stresses that a model-generated plan is only a proposal, so actions and results need validation. It also recommends stopping conditions such as step and time limits, with code handling exact arithmetic and rules. A research example illustrates how planning exposes intermediate steps and makes applications easier to debug.