AI glossary Models & Architecture
What is an agent planner in agentic AI?
← All glossary termsThe component or prompt pattern in an AI agent that breaks a goal into a sequence of steps before any tool is run.
Explained
An agent planner is the part of an AI agent that decides how to reach a goal: it breaks the request into steps, chooses which tools or sub-agents each step needs and proposes an order. Plans are usually conditioned on policies, specifications and retrieved context. The plan is then carried out by an executor, which calls the tools and reports results, and the planner may revise the plan as results arrive. Separating planning from execution makes agents easier to test and control: you can check a plan against acceptance criteria, permissions and budgets before anything with side effects happens.
How planning works in an agent
- Understand the goal and constraints: what is asked, what is allowed, what budget applies.
- Decompose: split the goal into steps small enough for one tool call or one sub-agent.
- Order and assign: decide the sequence and which tool, model or agent handles each step.
- Execute and observe: the executor runs the steps and returns results.
- Replan: if a step fails or reveals new information, the planner updates the remaining plan.
Common planning patterns
- Interleaved reasoning and acting: the model thinks, takes one action, observes the result and decides the next action. ReAct is the best-known example.
- Plan then execute: the model first writes a full plan, then works through it step by step, which makes the plan reviewable in advance. Plan-and-Solve prompting follows this idea.
- Orchestrator and workers: a planning agent delegates sub-tasks to specialised agents and combines their results.
Interleaved approaches adapt better to surprises; plan-first approaches are easier to inspect, approve and test.
Why separate planning from execution
A plan is a cheap, inspectable artefact: it has no side effects. Reviewing it before execution lets you check that the agent will only use permitted tools, stay within cost and time budgets and meet the acceptance criteria of the task. For sensitive workflows, a person can approve the plan, and the system can block execution of steps that were not in the approved plan.
Frequently asked questions
What is planning in AI agents?
It is the agent's ability to turn a goal into a sequence of actions, deciding which steps are needed, in what order and with which tools, and to adjust that sequence as it learns from results.
What is the difference between a planner and an executor?
The planner decides what should be done and in what order. The executor carries out each step, calling tools or sub-agents, and reports the results back so the planner can continue or revise the plan.
What is ReAct in AI agents?
ReAct is a prompting approach in which a language model alternates between reasoning about the task and taking actions, such as searching or calling a tool, and uses each observation to decide the next step.
Do all AI agents need a planner?
No. Simple agents with one or two tools can work step by step without an explicit plan. A separate planner becomes useful when tasks are long, involve several systems or need to be reviewed and approved before execution.
Where it shows up at Thinkia
Further reading
Related terms
Sources
- Yao et al. · ReAct: Synergizing Reasoning and Acting in Language Models (arXiv)
- Wang et al. · Plan-and-Solve Prompting (arXiv)
- Anthropic · Building effective agents
- Anthropic · How we built our multi-agent research system
Updated: 2026-10-21