Decomposition prompting
Breaking a complex task into smaller, simpler subtasks that a model can tackle independently.
Think of it like
Tackling a large project by breaking it into milestones.
Example
"Write an essay" → "Outline the thesis," "Write introduction," "Write body," "Write conclusion," "Proofread."
How it actually works
Decomposition reduces cognitive load on the model. Complex reasoning often fails end-to-end but succeeds when broken into steps. Downsides: more prompt calls (higher cost), error propagation (if step 1 is wrong, step 2 suffers). LLMs excel at planning decomposition; use one to plan, then execute.
For product teams
Improves reliability on hard tasks.
For engineers
Manually define decomposition or use an LLM to plan it. Implement error handling and recovery.
Related
- Prompt chaining — Executed via prompt chaining.
- Least-to-most prompting — Related to least-to-most prompting.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome