P321 · Agent orchestration

Differentiated design briefs

Generate alternatives under intentionally different constraints instead of multiplying identical proposals.

Editorially reviewed

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Explore file-import interfaces for one shared read/validate/result requirement. Teaching constraints differ: few entrypoints, format extensibility and easy common calls. Best-interface alone can produce near-identical proposals.

Mechanism

Vary optimization constraints while sharing domain terms/facts and output contract: interface, usage, hidden implementation, dependencies and tradeoffs. Compare actual behavior/shape, not names, and justify hybrids. One designer can explore sequentially; agent count is not diversity evidence.

Bad example

Request three best interfaces without different goals, then choose by naming/votes near the first idea.

Good example

Explore teaching A minimal entrypoints, B new-format extension and C easy single-file defaults. Require exact interfaces/invariants, usage, hidden work, dependencies and costs. Compare caller knowledge, extension and errors, not renamed functions. Explain recommendation/hybrid fit without assuming natural independence.

Why the change matters

Constraints expose different tradeoffs; common facts/formats make comparison valid. Diversity comes from actual interface consequences, not repeated identical prompts.

Observable expectation

An illustrative comparison distinguishes unified entry, registration and convenient defaults. Three renamed identical functions fail.

All retain shared validation/errors; omitted boundaries cannot manufacture flexibility. No real API is built.

Limits

Brief independence guarantees no thought independence. Counts/entry limits are source strategies rather than universal optima, and differentiated objectives cannot override common requirements.

Sources and evidence

Read the editorial criteria