What is a Task Decomposition?

A complex objective is rarely solvable in one step reliably; decomposition imposes structure by identifying the discrete pieces of work that together accomplish the goal, along with the dependencies between them (which pieces must finish before others can start, and which are independent and can run concurrently). Good decomposition finds the right granularity — sub-tasks specific enough to verify and execute reliably, but not so fine-grained that coordination overhead swamps the benefit. In agentic AI systems specifically, task decomposition is what an orchestrating agent (or a planning step within a single agent) does before dispatching work — identifying independent sub-tasks that can be delegated to sub-agents or tools, sequencing dependent steps correctly, and defining what a completed sub-task looks like so results can be verified rather than just trusted. Poor decomposition — either too coarse (one agent trying to do everything) or with hidden dependencies unaccounted for — is a common root cause of agentic systems producing inconsistent or incomplete results.