Course contentsShow
← Machine Learning and Deep Learning
Lesson 2085 of 3,53844. AI Agents: FoundationsFree lesson

Decomposition: Breaking Complex Tasks into Subtasks

Techniques for decomposing high-level goals into manageable subtasks that agents can execute sequentially.

Decomposition: Breaking Complex Tasks into Subtasks

What you'll learn: How AI agents transform overwhelming goals into manageable, sequential subtasks they can actually execute.

The Core Idea

Imagine asking an agent to "plan a vacation to Japan." That's too vague and multifaceted to execute in one step. Decomposition is the technique of breaking this down into concrete subtasks: research destinations, check flight prices, book accommodation, create an itinerary. Each subtask is smaller, well-defined, and actionable.

In agent systems, decomposition answers the question: "What smaller problems must I solve to achieve this bigger goal?"

Why Decomposition Matters

Complex tasks often require:

  • Multiple tool calls
  • Information from different sources
  • Sequential dependencies (you need flight dates before booking hotels)
  • Intermediate reasoning steps

Without decomposition, agents either fail completely or produce incoherent results because they're trying to tackle everything at once.

Common Decomposition Techniques

Sequential Decomposition: Break tasks into ordered steps.

  • "Analyze this dataset" → 1) Load data, 2) Check statistics, 3) Identify patterns, 4) Generate summary

Parallel Decomposition: Identify independent subtasks that can run simultaneously.

  • "Compare three products" → Query reviews for each product in parallel

Hierarchical Decomposition: Nested subtasks with multiple levels.

  • "Write a research report" → Outline sections → Research each section → Write drafts → Review and edit

Goal-Oriented Decomposition: Work backward from the desired outcome.

  • Goal: "Have presentation ready" → Need: slides, data, narrative → Need: raw data, analysis tools

The agent's planning module determines which subtasks to create, their order, and dependencies—then executes them through its perception-action loop.

Key Takeaway: Decomposition transforms abstract goals into concrete action sequences, enabling agents to solve complex problems through manageable steps that respect dependencies and maximize success rates.