The Planner Agent — From Requirement to Actionable Plan
How Datarim's Planner agent turns a vague requirement into a structured implementation plan with component breakdown, rollback strategy, and a concrete validation checklist.
A task description says "add export to CSV". That sentence contains no design, no rollback path, and no way to know when the work is done. The Planner agent takes that sentence and produces something a developer — human or AI — can actually execute.
The Planner is the Lead Project Manager of the pipeline. It runs at /dr-init to set up a task in the backlog, and again at /dr-plan to write the implementation plan. It uses the strongest model in the roster, because ambiguity in a plan compounds across every step that follows.
What it does
Every task gets a complexity level — L1 through L4 — that determines how much design work happens before implementation starts. A small bug fix is L1 and goes straight to code. A new service with a public API is L4 and gets a full design phase first.
For L3 and L4 tasks, the Planner produces a detailed design with three components: a component breakdown (what gets built and how pieces connect), an interface definition (what the code boundary looks like from the outside), and a data-flow diagram in prose. That design feeds a plan with explicit implementation steps, code examples, a rollback strategy using git or migration commands, and a validation checklist of specific checks — not «test it» but «run this command and verify this output».
One concrete example
A task arrives: add a webhook delivery system to an existing API. The Planner rates it L3. It reads the current codebase, identifies that a queue worker already exists, and writes a component breakdown showing three pieces — a queue producer in the existing request handler, a new worker process for delivery, and a retry table in the database. The implementation plan lists the migration command to add that table, the exact function signature for the producer, and the command to run to confirm delivery is working end-to-end. If something breaks mid-deploy, the rollback strategy is a single migration command that removes the table.
Where it sits
The Planner runs first in the pipeline — before any code is written. At /dr-init it registers the task and sets its complexity level. At /dr-plan it produces the plan document that every later agent reads. For complex tasks, the Planner's design feeds the Strategist, which evaluates whether the chosen approach is the right one before implementation begins.
See the full reference on the Planner agent card, or read what Datarim is for context on the full pipeline.