Goals, Tasks, and Decomposition
From one goal to many steps
Why decomposition matters
Many real-world goals are too large to solve in one step. Agentic AI often breaks a task into smaller subtasks. This is called task decomposition. For example, "organize a study plan for biology" might become: identify topics, estimate available time, schedule sessions, and review progress.
Rule of thumb
If a goal has multiple dependencies, an agent usually performs better when it plans before acting.
Example decomposition
| Big goal | Smaller tasks |
|---|---|
| Write a report | Collect sources, outline sections, draft, revise |
| Book a trip | Set budget, compare options, check dates, confirm booking |
| Learn a topic | Find basics, practice, test understanding, review mistakes |
Planning styles
Some agents plan the whole path up front. Others plan only one step at a time and adjust as new information appears. Both approaches can work depending on uncertainty, time, and risk.
What is task decomposition?
Decomposition means splitting a complex goal into smaller pieces.
Correct answer: Breaking a big goal into smaller steps
Why do agents often need to plan before acting?
Planning helps the agent sequence those steps correctly.
Correct answer: Because complex tasks usually require multiple dependent steps.
A beginner-friendly mental model
Think of an agent like a careful assistant: it listens to the objective, makes a checklist, carries out the checklist, and updates the checklist when conditions change.
GOAL DECOMPOSITION
How agents break big goals into manageable steps
From Goal to Action: A Tree View
Clear success criterion — e.g. "Return a JSON list of 10 product names."
No clear end state — e.g. "Make the website better." Agent must infer subtasks.
Goals that spawn more goals. Common in complex research or multi-agent setups.