Task Decomposition

Task decomposition is the practice of taking something complex — “analyze this document and write a strategic recommendation based on it” — and breaking it into discrete steps: first summarize the key points, then identify the central tensions, then draft a recommendation with supporting arguments. This approach is more reliable because each subtask is simpler, the outputs of each step can be checked, and errors in one step don’t automatically propagate unnoticed to the final output. Task decomposition applies to how you design prompts (chain-of-thought, prompt chaining) and how you structure multi-agent workflows. For behavior architects, decomposing tasks explicitly also produces more interpretable results — when a model fails, you can identify exactly which step went wrong.