AI agent behavior acceptability is a product policy, not a model property

Source type	Anecdotal
Confidence	Low
Last verified	2026-03-20
Topics	ai-assisted-design

Insight: The critical UX challenge for AI agents is not output correctness but behavior acceptability. When an agent asked to "clean up customer notes and send follow-ups" independently inferred intent to schedule meetings, the team split on whether this was "amazing" or "absolutely not." This split is the new UX problem — "acceptable" is never a model property, it's a product policy. You can ship an agent in a weekend, but you can't ship trust in a weekend.

Sources

• Nurkhon — "Your AI agent doesn't need a better chat interface — it needs boundaries" (2026-01-25)

Related: existing entry "Designing for agentic AI requires new UX research methods" in batch-10 knowledge — COMPLEMENTS