Enterprise AI assistant
An assistant that tells people plainly what it can see, what it can do, and where every answer came from.
Why it mattered
Staff were pasting internal content into public chatbots. A model is only useful inside an organization if it knows the terminology, respects permissions and behaves predictably. A confident wrong answer about a policy is worse than no answer.
Who we designed for
The writer
Drafts announcements and emails daily. Wants a starting point and needs to know which internal source an answer came from.
The skeptic
Has seen AI invent facts. Will only use a tool that shows its sources and admits what it cannot do.
The power user
Uses it for code and document analysis. Wants history, export and keyboard speed, without hand-holding.
Where it got stuck
Not knowing what the assistant had access to, so people over- or under-trusted it.
Answers with no provenance.
Hitting a usage limit with no warning and no idea when it would reset.
Losing a good conversation because there was no history.
The decisions
Say what it can see
Onboarding states exactly which data the assistant can access. That one line did more for trust than any visual polish.
Starter prompts as examples
Cards show complete, realistic requests, and double as one-tap actions.
Every answer cites
Reference chips link to the documents used, so the output stops being a black box.
Limits are told, not discovered
An inline message explains a cap and when it resets, in the compose box, before the user types.
Four alert tones, one shape
Danger, caution, success and general share one layout and differ by color and icon, in both themes.
Compose box as the control center
Attach, regenerate and send live in one bar, with tool menus in a consistent position on web and mobile.