Principle 08
Minimal Ambiguity
Each term, action, and value should have one meaning. When a request could mean more than one thing, the product should ask rather than guess.
Why it matters
Ambiguity compounds in agentic systems. A person asks an agent for "the cheapest flight"; the agent interprets "cheapest" one way; the product interprets the agent's query another. Each interpretation is reasonable, and the result is wrong.
Reducing ambiguity at the product layer is cheaper than asking every agent to resolve it independently.
Examples
Good exampleIllustrative
A search for "Paris" returns a clarification listing Paris, France and Paris, Texas, each with an identifier, instead of silently choosing one.
{
"status": "needs_clarification",
"field": "destination",
"options": [
{ "id": "CDG-PAR", "label": "Paris, France (all airports)" },
{ "id": "PRX", "label": "Paris, Texas, United States" }
]
}Bad exampleIllustrative
"Delete" archives items on one page and permanently removes them on another.
Implementation guidance
- 1Publish a glossary for domain terms and use it in data, UI, and docs.
- 2Use enumerations instead of free text for values with a fixed set of options.
- 3Return clarification responses for ambiguous inputs.
Evaluation criteria
- The same term means the same thing everywhere it appears.
- Ambiguous inputs produce a clarification, not a guess.
- Enumerated fields list their allowed values.