FollowAgents AI AGENTS DIRECTORY
Home / AI Agent Glossary / Task Decomposition
Skills & Workflow

Task Decomposition

Task decomposition is the process of breaking one large or ambiguous task into smaller, more concrete subtasks that can each be executed and checked individually.

Subtask 1Subtask 2Subtask 3Big task
One big task splits into several independent subtasks.

“Add authentication to this app” is a fine request for a human, but it's too large and underspecified for an agent to just execute in one shot. Task decomposition is the step where an agent (or a person directing one) breaks that big task into smaller pieces it can actually act on: design the schema, add the login endpoint, add session handling, write tests.

Decomposition matters because smaller subtasks are easier to get right, easier to verify individually, and easier to hand off — to a different subagent, to a different tool, or to a later step in a workflow — than one large, vague instruction. It's also what makes a multi-agent-system practical: you can't usefully split work across agents without first breaking it into pieces.

There's no single required structure for decomposition. It's often drawn as a tree or hub-and-spoke (one big task, several subtasks), but a flat list of subtasks with no hierarchy at all is just as legitimate a decomposition, depending on whether the subtasks actually depend on each other.

Example

Given “add authentication to this app,” an agent might decompose it into: add a users table and password hashing, add login/logout endpoints, add session or token middleware, and add tests covering both successful and failed logins.

How it differs

Task decomposition vs. planning: decomposition specifically means splitting a task into subtasks; planning is the broader activity of deciding what to do, which usually includes decomposition plus figuring out the order and dependencies between the pieces.

Common misconceptions

Often assumed: Decomposition always produces a strict tree of subtasks.
Actually: It's often drawn that way, but a flat, unordered list of subtasks is a perfectly valid decomposition when the pieces don't actually depend on each other.

FAQ

What is task decomposition in AI agents?
Breaking one large or vague task into smaller, concrete subtasks that can each be worked on and checked individually.
How is task decomposition different from planning?
Decomposition is specifically about splitting a task into pieces; planning is the broader process that also figures out the order and dependencies between those pieces.

Last checked: 2026-08-28

Related terms

Planning

Planning is the step where an agent turns a goal into an ordered set of smaller steps before — or while — it starts acting, so it isn't deciding what to do purely one action at a time with no larger structure.

Chain-of-Thought (CoT)

Chain-of-thought is a prompting technique where a model is asked to write out its intermediate reasoning steps before giving a final answer, rather than producing the answer directly.

Multi-Agent System

A multi-agent system is a setup where multiple AI agents, each with a narrower role, work on parts of a larger task and coordinate through some form of orchestration.