AI can play different roles during software work. The failure mode is treating every task as if it wants the same level of rigor.
These four modes are a practical way to choose the right “contract” with the model for the moment.
1) Brainstorming
Use when you want options, framing, and contrast, not correctness.
- Goal: expand the space of ideas
- Output: rough lists, questions, alternatives
- Risks: plausible nonsense; hidden assumptions
2) Drafting
Use when you want a first version you will edit (text, code, docs).
- Goal: reduce blank-page cost
- Output: a coherent draft you can shape
- Risks: overconfident tone; missed edge cases
3) Pairing
Use when you want a thinking partner while you stay responsible for decisions.
- Goal: faster iteration with checks
- Output: small deltas, tight feedback loops
- Risks: context drift; “fixing” the wrong thing
4) Surgical changes
Use when the change must be minimal, reviewable, and safe.
- Goal: precise edits with clear boundaries
- Output: targeted patches, diffs, test updates
- Risks: missing broader implications; local optimums
Choosing quickly
If you can’t articulate the mode, default to pairing: small steps, frequent verification, explicit assumptions.