Decomposition & Prompt Chaining

By the end you will turn a big task into a reliable chain of prompts.

See it first

Ask AI to "research my competitors, write a positioning statement, and turn it into a launch email" in one prompt, and you get a rushed, shallow version of all three. Split it into steps, where each step's output feeds the next, and each part gets the model's full attention.

This is prompt chaining: decompose the task, then run a sequence where the output of step N becomes the input of step N+1. It is the same "break the work into steps" idea from the last lesson, made into a repeatable pipeline.

When one prompt is not enough

Reach for a chain when a task has distinct phases, when later steps depend on earlier results, or when one prompt would force the model to juggle too much at once. Quality drops when a single prompt has to do research and analysis and writing and formatting together.

Passing output forward

The core move: take what step one produced and feed it into step two. A typical workflow:

Research: "List the top 5 competitors for X with their positioning."
Outline: "Using that list, outline how we should differentiate."
Draft: "Write a positioning statement from that outline."
Critique: "Critique that statement as a skeptical customer."
Polish: "Revise it addressing those critiques, in a confident tone."

Each step is small, focused, and checkable. You carry the best output of each into the next, fixing problems before they compound.

What you can now do

Recognize when a task needs a chain instead of a single prompt
Decompose a task into focused, sequential steps
Pass the output of one step into the next to build a workflow
Place human review at the right handoffs to stop errors compounding

Check your understanding

1 / 2

prompt-chaining

What is prompt chaining?

What's next

You can structure the work. Now sharpen each individual prompt with the core toolkit: Design of a Prompt: Nine Strategies.