A feature that lets users describe what they want in natural language sounds like the modern, AI-native answer. For dates, addresses, and structured choices, it is often the answer that actively annoys the people you most want to retain. The UX principle most AI teams skip — and what changes when you apply it.
Letting users describe everything in natural language sounds AI-native and quietly punishes power users with structured intents. The principle: modality belongs to the data, not the brand. Native controls for finite-shape inputs (dates, addresses, numbers, choices); chat for genuinely open-ended intent. Hybrid surfaces win on both quality and cost.
You are wrong if you make every input chat-driven. Power users with structured intents — a known date, a specific entity, a fixed quantity — actively prefer native controls. Chat as the default surface for structured data is brand-driven design, not user-driven. The healthy pattern is hybrid: chat for ambiguity and exploration, native controls for everything else.
Time-to-confirmed-input — how long it takes a user to provide a piece of structured data and confirm the system understood it correctly. For power users, NL inputs frequently take 3×–10× longer than equivalent native controls. CSAT and engagement lag this by weeks; time-to-confirmed-input moves first.
No, the opposite. Removing chat as the wrapper around structured data lets the model do what it is uniquely good at — handling open-ended intent, ambiguity, exploration — while native UI handles what it has always been better at. Cost falls, quality rises, and the product feels more competent, not less AI-native.