The approval interface — a new interaction pattern where users evaluate outputs rather than construct inputs¶
Insight: According to Felix Haas, a new interaction pattern is emerging across AI-powered software: the "approval interface," where "the user's job is less and less about constructing inputs and more and more about evaluating outputs." The loop: minimal user input, AI generates result, AI proposes next action proactively, user approves or redirects, loop repeats. Products using this pattern include Lovable, Cursor, ChatGPT, and Notion AI.
Detail: Haas traces the pattern's origin to Pinterest's guided search, where users typed one word ("living room") and received suggested refinements ("Cozy," "Minimalist," "Scandinavian"). The modern version represents a shift "from initiating to approving" — the interface runs "slightly ahead of you," requiring users to actively evaluate whether to accept, modify, or reject suggestions. This compression of the interaction loop reduces friction while maintaining user agency through approval mechanisms. The pattern has direct implications for design system tooling where AI generates components and users approve or iterate.
Sources
- Felix Haas / Design + AI — "The Approval Interface" (2026-02-19)
Related: agent-ui-progressive-disclosure-patterns (this batch) — COMPLEMENTS