An AI-led async alignment tool inside Arthur's enterprise workspace. Set up a Vibe, invite participants, an AI bot interviews each person, a report is generated.

Used by enterprise clients including PwC, Mondelez, and Bayer.

Move recurring meetings out of the calendar and into a structured async interview.

The challenge was not the interview. It was making the output trustworthy enough that managers would actually skip the meeting.

I worked on the product with another designer, splitting screens across the flow.

Most async meeting tools fail for the same reason. The output does not earn the manager's confidence. Summaries get skimmed once and ignored.

For Vibe to work, the report had to feel like a decision-quality artifact. Every design decision downstream traced back to this.

The interview had to feel low-friction, natural, and honest. Small moments where trust could break got the most attention.

One line to set expectations: who invited them, what the Vibe is about, how long it takes. No signup, no dashboard.

Every answer is editable in the review step. Once submitted, responses lock and the report generates.

Async lives and dies on completion. So the email system was designed as its own product.

Every Vibe triggers conditional emails based on participant behavior. Did not open, did not finish, deadline approaching. Each state gets its own message.

Enterprise Vibes often include people outside the client's account. Making them create an account for three questions would kill completion.

A shareable Vibe URL lets anyone with the link answer. Optional email confirmation for identity. The tradeoff (weaker verification, not fully SOC2 compliant) was documented in the design file and surfaced to the team before shipping.

Vibe launched as a new tab inside Arthur's portal, running today for teams already using the platform.

Analytics were not in scope. What the design guarantees: async interviews across multiple meeting types, a participant flow that works across desktop and browser, an email system that adapts to behavior, and access controls that meet enterprise expectations.

Every decision, from the welcome screen to email triggers to access controls, was the same decision at a different scale. Does this make the output worth trusting when a manager has to act on it in five minutes?

The hardest part of designing AI products is not the AI. It is designing the surface around it so the output earns its place in the workflow.