Context
Archie is an AI-native accounting platform. The goal was to turn complex accounting operations into structured digital experiences that accountants can trust, control and audit.
The problem
Accountants work under professional responsibility. An AI that acts without showing its reasoning is unusable for them, however fast it is. The design question was how AI should behave in a professional product while keeping human oversight and transparency.
Accounting processes involve large chunks of work and a lot of documentation, much of it never automated before. Chasing clients for information is complex, and an audit has to be perfect.
What I did
I defined the product concepts and the experience around them: Topics, WorkStreams, Cycles, Archie's Inbox and The Chase. For the V3 platform I rethought the product around Topics as the central collaboration space, bringing together emails, documents, agent actions, client communications and workflows.
I also created Archie's design system and enterprise product language. The work covered web, desktop, an Excel sidebar, a Teams panel and operator-style AI workflows.
Deliverables: PRDs, UX specs, prototypes, interface explorations and design prompts for implementation.
How I worked
- Wrote PRDs and UX specs so product and engineering shared one definition before UI work.
- Explored interfaces and prototyped the flows: document collection, client communication, workflow execution and accountant approvals.
- Built the design system as components documented in Storybook. I implemented the code and handled the handoff to development. We used Chromatic to version Storybook, and I applied the system to real flows with Claude Code.
Key decisions
Make AI work visible, not hidden.
I designed a Chain of Evidence so an accountant can see what supports each action. Showing only the final result was not enough for users who need a clear audit trail of how things were done.
Balance automation with human oversight.
Approval flows, progress visibility and audit trails keep the accountant in control.
Topics as the center of the product.
One shared space instead of separate tools for email, documents and agents. It simplifies the work instead of adding a trade-off: everything about a topic lives in one place, so users can check communications, see work in progress and know what needs their immediate attention.
A dense, persistent-navigation desktop app.
A status-rich enterprise interface suited to people who spend the day in it.
Outcome
The product reached the market stage where it could be tested with real customers: several pilots were signed with enterprise companies. I left before results came in, so I can speak to the design, the system behind it and the path to those pilots, not to long-term metrics.
What I took from it
As the only designer, I moved between product, commercial work and code, and learned to ship the whole thing instead of handing off a file. The lesson that stayed with me: in a professional product, a good result is not enough. People need to see how the AI got there. Designing for AI means designing that trail, and a design system engineers can use from day one.