Slapping a chat panel onto a product and calling it an AI feature is the interface equivalent of shrugging. A blank text box outsources the hardest part of the interaction — figuring out what to ask, in what words, with what context — to the person who came to you precisely because they didn't want to figure that out.
Why chat became the default
Chat is what the underlying technology looks like, so it's the cheapest thing to expose. It also demos beautifully: an open-ended box looks infinitely capable on stage. In real usage the same openness becomes a cold start problem — most users try two generic prompts, get mediocre results, and never come back.
Design for outcomes, not conversations
The better pattern is to bury the model inside verbs the user already understands: a button that drafts the reply, a default that's already personalized, a suggestion that appears exactly where the next action happens. Equally important is designing for verification — the user must be able to confirm at a glance that the AI got it right, and fix it in one gesture when it didn't.
We watched this play out with a fitness training app. The original plan was an AI coach you chat with. In testing, users had nothing to say to it. What worked was invisible: after each workout the app drafts the next session automatically, with one tap to make it easier or harder. Same model underneath — but now it does a job instead of holding a conversation.
The verdict
Nobody opens your product hoping to chat with it. They open it to finish something. If your AI feature can't be described as a verb with an outcome, it's not a feature yet — it's a demo.
