Most AI design stops at the chat window. That's the easy part.

The hard problems start when software acts on someone's behalf: how a person hands off work and still feels in control, what the agent shows while it's thinking, and what happens when it's confidently wrong.

I've been designing that since 2022: an agent inside a team workspace, an agent that spends your money, an agent that judges your work. Two shipped into production platforms. One is my own product, designed and built solo.

I work on AI-native SaaS: agent interfaces, dashboards, and the design systems underneath them. If your product does something on a user's behalf and you need them to trust it, follow it, and be able to stop it. That's the work.

Design thesis. Before pixels: one direction, argued. What I'd build, what I'd cut, and why. You approve the thinking before I spend your budget on execution.

Flows and screens. User flows through high-fidelity UI. Not just the happy path, but the states where the agent is uncertain, wrong, or waiting on you. That's where AI products actually fail.

Design system. Tokens, components, variants, documentation. Built so your team can extend it without me.

Figma source + handoff. Editable files, plus implementation notes. I write front-end (HTML/CSS/Tailwind), so my specs get built the way they're drawn.