Agentic Toolkit Course
Lesson 3 of 7

Lesson 3 of 7 · Individual agent

The Planner agent

In software development we're often making architectural decisions and want to figure out how something can be done before implementing anything. Those structural decisions are exactly what the Planner helps with. The Planner understands the current code and gives structured ways of how something can be done, including edge cases, risks, open questions, and how the implementation can be validated. The Planner never writes code — the Planner only describes what needs to happen.

When to use the Planner

How to use the Planner

  1. Describe the concrete requirement or task — not just "make it better".
  2. Name the files or areas of the codebase you think are involved.
  3. State constraints: technology, existing patterns, backward compatibility.
  4. Describe what the end result should look like, and list any open questions.
  5. Review the plan the Planner returns — the plan should include a summary, standards discovered, verified files/symbols, ordered steps, edge cases, risks, and a validation strategy.
"I need to add role-based access control to our Express API. Currently
all routes are open. I want admin, editor, and viewer roles. The routes
are in src/routes/ and the auth middleware is in src/middleware/auth.ts.
How should I approach this? What are the edge cases?"
Where the Planner fits in the Orchestrator: the Planner runs as Stage 1. The Orchestrator verifies the Planner's plan (checking file paths and symbols are real) before presenting the plan for your approval and passing the plan downstream to the Designer and Coder.

Things to avoid

Test your knowledge

You want to evaluate three different ways to migrate a monolith to microservices before writing anything. Which agent?

Source: documentation.md, "Planner".

← Back: Orchestrator flow Next: Designer agent →