ModelRefs / Plan-and-Execute — Agent Pattern
Plan-and-Execute — Agent Pattern
Split the agent into a planner that drafts a multi-step plan and an executor that runs each step with tools.
Overview
A high-context planner LLM decomposes the task into ordered steps. An executor LLM runs each step with tools, optionally replanning when steps fail. Separates reasoning from action for cheaper, more debuggable runs.
When to use it: ReAct loops are too greedy and you need explicit, inspectable plans before any tool runs.
Pattern details
| Pattern class | planning |
|---|---|
| Difficulty | intermediate |
| Autonomy | guided |
| Also known as | planner-executor, two-agent planner |
| Last reviewed | 2026-06-07 |
Known failure modes
- Plan drift — Executor deviates silently from the plan. Mitigation: Constrain executor to one step at a time; log adherence.
- Brittle plan — Plan fails at step 2 with no recovery. Mitigation: Add a replanner with explicit retry budget.
When not to use it
- Planning for tasks ReAct solves in two steps.
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Plan-and-Execute — Agent Pattern.
Frequently asked questions
When should I use the Plan-and-Execute agent pattern?
ReAct loops are too greedy and you need explicit, inspectable plans before any tool runs.
What are common failure modes of Plan-and-Execute?
Plan drift • Brittle plan
Is Plan-and-Execute production-ready?
Yes when paired with the safety controls and observability hooks documented on the pattern page.