# What is Task Decomposition?

Also called problem decomposition, subtasking.

Task decomposition is breaking a large or vague request into smaller, concrete subtasks that can be executed and checked one at a time. In agent systems the split may be written by a developer, produced by the model at runtime, or a mix of both. Good decomposition produces subtasks with clear inputs, clear completion criteria, and few dependencies.

Decomposition addresses two limits at once. Long tasks exceed what a model can hold and track in a single pass, and undefined success criteria make failure impossible to detect. Splitting the work gives each unit a small context and a testable outcome, so an error surfaces at the step that caused it rather than at the end of a long run.

The split can be fixed or generated. A fixed decomposition, written once as workflow stages, is repeatable and easy to test but only fits work with a known shape. A generated decomposition lets the model produce a plan for the specific request, which handles novelty at the cost of variability. Systems often fix the top level and generate below it.

Common failures are recognizable. Subtasks too large to verify hide errors. Subtasks too small multiply overhead and lose the thread. Hidden dependencies cause a later step to need something an earlier one never produced. And plans generated up front go stale, because step two often reveals that the original step five was the wrong thing to do.

Decomposition is the mechanism behind planning, delegation, and multi-agent design. A supervisor agent decomposes and hands pieces out. A workflow encodes a decomposition someone already made. Replanning, where the agent revises the remaining steps after each result, is the standard answer to plans that go stale during execution.

## Key points

- Splits vague work into subtasks with clear inputs and endpoints.
- Decomposition can be fixed in code or generated per request.
- Oversized subtasks hide errors; undersized ones add overhead.
- Plans made up front go stale and need revising mid-run.

## In practice

The request is to prepare for a customer renewal call. An agent breaks it into four pieces: pull the account's usage over the last quarter, list every support ticket they opened, check which contracted features they never enabled, and summarize the three into talking points. Each piece has an obvious finish line, and a failure in any one is visible immediately.

## Related terms

- [Agent Planning](/en/glossary/agent-planning)
- [Delegation](/en/glossary/delegation)
- [Supervisor Agent](/en/glossary/supervisor-agent)
- [Agentic Workflow](/en/glossary/agentic-workflow)
- [Multi-Agent System](/en/glossary/multi-agent-system)

[Back to the AI Glossary](/en/glossary)