Product design for AI tools, where the hard problem is trust rather than layout.

Who this is for. Teams building an AI product: an assistant, an agent, a copilot, a data platform with a model underneath it. You have the model working and the interface is where it stops being convincing.

Why AI products need different design. A chat box is not a product. The real problems are specific: how a user starts when they do not know what to ask, how the system shows its reasoning without drowning them, what happens when the model is wrong, and how confidence gets communicated without either overclaiming or undermining the tool. I have solved these on an AI legal assistant handling sensitive case work, an AI learning platform serving three different user roles, and an AI market intelligence platform for ticketing brokers.

What I do about the blank prompt. In Emmi, the AI legal assistant, users start from labelled practice areas rather than an empty input. That one structural decision removes the blank page problem and gives the model its context before the first message. Every AI product needs its own version of that move.

Process. Discovery and flows, then core screens, then the design system, then handoff built for engineers. You review in Figma throughout.

Scope. Eight to twelve weeks depending on surface count. The starting price covers a focused scope, agreed in writing before we begin.