Decompose a large, fuzzy task into an ordered sequence of smaller prompts and assemble the result.
Why one mega-prompt fails — and how to decompose a big task
Why one mega-prompt fails — and how to decompose a big task
When you throw a large, complex task at the model in a single prompt — 'Write me a complete 20-page onboarding guide for new sales reps' — several things go wrong at once. The model must invent structure, tone, depth, and content simultaneously, with no checkpoints for you to steer. The result is often generic, uneven, and hard to fix in one pass. Large prompts also risk filling the context window, which can cause the model to lose track of earlier instructions.
Task decomposition means breaking the big task into an ordered sequence of smaller prompts, where each step produces a clear artifact that feeds the next. A typical decomposition looks like: (1) Outline — ask the model to propose a structure and review it yourself; (2) Draft each section — one prompt per section; (3) Tighten — ask the model to shorten and improve each section; (4) Format — ask the model to assemble the final document in the required format. This is sometimes called prompt chaining: the output of one prompt becomes the input of the next.
The key benefit is control: you can review and correct after each step before moving on, which means errors stay small and the final result matches your actual needs. You are not outsourcing the whole task — you remain the editor and decision-maker throughout.
Worked example: You need a report on why Q2 sales missed target. Step 1 prompt: 'List the main sections a 2-page root-cause report on a missed sales target should include. Return a numbered outline only.' Step 2 prompt: 'Using this outline [paste outline], draft the Executive Summary section in 100 words, formal tone.' Step 3 prompt: 'Make this Executive Summary 20% shorter without losing key facts.' Step 4 prompt: 'Combine all sections into a single formatted document with bold section headers.'
Lesson notes
Why one mega-prompt fails — and how to decompose a big task
When you throw a large, complex task at the model in a single prompt — 'Write me a complete 20-page onboarding guide for new sales reps' — several things go wrong at once. The model must invent structure, tone, depth, and content simultaneously, with no checkpoints for you to steer. The result is often generic, uneven, and hard to fix in one pass. Large prompts also risk filling the context window, which can cause the model to lose track of earlier instructions.
Task decomposition means breaking the big task into an ordered sequence of smaller prompts, where each step produces a clear artifact that feeds the next. A typical decomposition looks like: (1) Outline — ask the model to propose a structure and review it yourself; (2) Draft each section — one prompt per section; (3) Tighten — ask the model to shorten and improve each section; (4) Format — ask the model to assemble the final document in the required format. This is sometimes called prompt chaining: the output of one prompt becomes the input of the next.
The key benefit is control: you can review and correct after each step before moving on, which means errors stay small and the final result matches your actual needs. You are not outsourcing the whole task — you remain the editor and decision-maker throughout.
Worked example: You need a report on why Q2 sales missed target. Step 1 prompt: 'List the main sections a 2-page root-cause report on a missed sales target should include. Return a numbered outline only.' Step 2 prompt: 'Using this outline [paste outline], draft the Executive Summary section in 100 words, formal tone.' Step 3 prompt: 'Make this Executive Summary 20% shorter without losing key facts.' Step 4 prompt: 'Combine all sections into a single formatted document with bold section headers.'