- Every effective prompt draws on the same components: role/context, task, background data, rules/tone, examples, output format — specificity means precision on audience/outcome/format/length, not length of the prompt itself
- Decompose a task when it has a real pipeline shape or needs an inspectable intermediate step; don't decompose what a single coherent pass already handles
- A weak output is diagnostic — trace it back to the thin structural component (missing context, under-specified instruction, or missing example) and fix that, rather than just retrying
- Task type changes which components matter most: analysis leans on rules/criteria, research on background data and sourcing, drafting on examples/tone, brainstorming on volume with deferred judgment
Key Takeaways
Estimated time: 3 minutes