Learn / Workflow Engineering / Task Decomposition

Task Decomposition

10 min apply 4 sections
Step 1 of 4

WHY WHY This Matters

AI excels at focused, well-defined tasks. It struggles with ambiguous, multi-part requests. The skill of breaking complex work into AI-manageable chunks is what separates effective operators from frustrated users who complain "AI doesn't understand what I want."

The decomposition advantage:

  • Higher quality outputs (focused attention)
  • Better error isolation (know which step failed)
  • Reusable components (build a library of prompts)
  • Clearer human oversight (review at checkpoints)
Step 2 of 4

WHAT WHAT You Need to Know

The Atomic Task Principle

Decomposition example: "Write a marketing campaign"

❌ Too big (not atomic):

"Create a complete marketing campaign for our product launch"

✅ Properly decomposed:

  1. Generate 10 campaign theme concepts (brainstorm)
  2. Evaluate themes against brand guidelines (analysis)
  3. Develop messaging pillars for chosen theme (strategy)
  4. Write email subject lines (copy generation)
  5. Write email body copy (copy generation)
  6. Create social media post variants (adaptation)
  7. Generate ad headlines (copy generation)
  8. Review all copy for consistency (quality check)

Dependency Mapping

Dependency visualization:

[Research] → [Outline] → [Draft] → [Edit] → [Final]
     ↓           ↓
[Competitor  [Examples
 Analysis]   Search]
     ↓           ↓
     └─────→ [Synthesize] ─→ [Insert into Draft]

The Orchestration Pattern

Complex workflows often follow this pattern:

┌─────────────────────────────────────────────────┐
│                 ORCHESTRATOR                     │
│  (coordinates tasks, manages state, handles      │
│   errors, combines outputs)                      │
└──────────────┬──────────────────────────────────┘
               │
    ┌──────────┼──────────┐
    ↓          ↓          ↓
[Task A]   [Task B]   [Task C]
    ↓          ↓          ↓
[Output A] [Output B] [Output C]
    └──────────┼──────────┘
               ↓
        [Combined Result]

Orchestration responsibilities:

  • Task sequencing and parallel execution
  • Passing outputs between tasks
  • Error handling and retry logic
  • Human checkpoint management
  • Final assembly of outputs

The Decomposition Method

Step 1: Start with the end What's the final deliverable? Work backwards.

Step 2: Identify natural breaks Where would you naturally pause, review, or make a decision?

Step 3: Define inputs and outputs For each chunk, what does it need? What does it produce?

Step 4: Check atomicity Can each chunk be done by AI without asking clarifying questions?

Step 5: Map dependencies Which chunks depend on others? Which can run in parallel?

Step 6: Add checkpoints Where should humans verify before proceeding?

Key Concepts

Key Concept

atomic task

An atomic task is the smallest unit of work that produces a meaningful, verifiable output. For AI, atomic tasks should:

  • Have a single, clear objective
  • Require no mid-task decisions
  • Produce output that can be verified independently
  • Take seconds to minutes, not hours
  • Be repeatable with consistent results
Key Concept

dependencies

Dependencies are relationships between tasks where one must complete before another can start. Types include:

  • Sequential: Task B needs output from Task A
  • Parallel: Tasks can run simultaneously
  • Conditional: Task B only runs if Task A produces specific result
  • Iterative: Task repeats until condition met
Step 3 of 4

HOW HOW to Apply This

Exercise: Decompose Complex Tasks

Task Decomposition Template

# Task Name Input Output Dependencies Human Checkpoint?
1 None
2 Step 1
3 Steps 1, 2
4 Step 3 ✓

Common Decomposition Patterns

Complex Task Decomposition Pattern
Research report Gather → Filter → Analyze → Outline → Draft → Cite → Review
Content creation Brief → Research → Outline → Draft → Edit → Format → Review
Data analysis Extract → Clean → Transform → Analyze → Visualize → Narrate
Decision support Frame question → Gather options → Evaluate → Pros/cons → Recommend
Communication Audience analysis → Key messages → Draft → Tone check → Finalize

When to Combine vs. Separate

Combine When Separate When
Tasks share context that's expensive to repeat Tasks might have different quality needs
Output of one is trivially small Mid-point review is valuable
Combined takes < 30 seconds Either task might fail independently
No human judgment needed between Different expertise/models optimal

Self-Check

Practice Exercises

Task 1: Quarterly Business Review Presentation Break into steps covering: data gathering, analysis, story creation, slide content, and review.

Task 2: Customer Complaint Resolution Include: understanding the issue, researching history, drafting response, and escalation logic.

Task 3: Job Posting Creation Cover: requirements gathering, competitive analysis, writing, compliance check, and formatting.

For each decomposition, identify:

  • Input needed for each step
  • Output produced by each step
  • Dependencies between steps
  • Where human checkpoints should be
Step 4 of 4

GENERIC Up Next

In Module 2.3: Quality and Iteration, you'll learn how to build quality assurance into AI workflows—validation, iteration, and feedback loops.

Module Complete!

You've reached the end of this module. Review the checklist below to ensure you've understood the key concepts.

Progress Checklist

0/5
0% Complete
0/4 Sections
0/2 Concepts
0/1 Exercises