Patterns
The moments that decide whether people trust an AI product, and what to do about each. Open one to try it.
- 01Ask before actingThe assistant wants to do something that cannot be undone. What now?Stop and ask first. Say exactly what will happen, then wait for a clear yes or no.
- 02Show your sourcesThe assistant states a fact. How does a person know it is true?Put a small numbered mark next to each claim, and list where it came from.
- 03Admit uncertaintyThe assistant is not sure of its answer. How should it say so?Say it plainly, show how sure it is, and tell the person what to check.
- 04Show work as it happensThe assistant takes a while. What does the person see in the meantime?Show each step as it happens, so waiting feels like progress and not a stall.
- 05Let people steerThe assistant is heading the wrong way. How can a person change course?Give them Stop, Regenerate, edit and a way to go back, always within reach.
- 06Recover from mistakesSomething failed, or the assistant got it wrong. What should the screen say?Say what happened, say that the person's work is safe, and offer a clear next step.
- 07Make AI content obviousA person is reading text. Can they tell who wrote it, the assistant or a human?Mark what the assistant wrote, mark what a person changed, and let people go back.
- 08Respect attentionMany things are happening. Which ones should interrupt a person?Only the ones that need a decision. Keep everything else quiet and easy to find later.
- 09Watch the screenThe assistant is using a website for you. How do you know what it is doing?Show its screen with a line for the current step. Let people pause or take over, and ask before sensitive steps.
- 10Share the roomSeveral people and the assistant are in one chat. Who is it talking to, and what can it see?Show who is speaking, let the assistant speak only when asked, and say what it can see.
- 11Mark the areaA person wants to change one part of a picture or page. How do they say which part?Let them mark the area, say the change, then compare before and after, with Undo close by.
- 12Let the agent askThe assistant is unsure in the middle of a task. Should it guess?No. It asks one clear question with a few choices, and the person can let it decide.
- 13Come back after time awayA person returns after the assistant worked alone. What do they see first?A short summary: what finished, what needs them, what changed, and one tap to the full log.
- 14Set how much it can doHow much should the assistant do on its own, and who decides?Start careful, let people raise it, confirm only what cannot be undone, and keep the rest visible and undoable.
- 15Generate screens safelyThe assistant wants to build a screen for the person. How do you keep that safe?The assistant picks parts from an approved list. The app draws them in its own look.