Chat foundations
People and companies rendered as distinct objects, a standard template for tool calls, and a composer where mentions resolve to real records.
For a platform where agents build a new app for every customer, we designed the component library and wrote the design rules the agents follow.

An AI-native intelligence platform in stealth. Customers search, track and act on data through chat, automations and an agent. Each customer gets its own app, built by agents and engineers, and every app pulls from one shared design system. One component change reaches every deployment.
The product worked, but the experience felt chaotic and the visuals undersold it. The team needed a library that engineers without design training could use to build polished front ends for every customer.
The design system has a new reader. Write it for the agent too.
Chat foundations
People and companies rendered as distinct objects, a standard template for tool calls, and a composer where mentions resolve to real records.
A home portal
Draggable widgets for active work and pending actions.
Three detail patterns
A side drawer for browsing, an overlay for focused tasks, and a full-page expand.
The component library
Forms, tags, data tables, chat bubbles, thinking states, action cards, empty and loading states, toasts, drawers and charts, built in Figma and React.
Per-customer theming
The same system re-skins for each customer.
Guidance agents can follow
Motion and interaction rules written into each component's docs, so agents building new pages choose the right pattern on their own.




Four weeks. We polished a few key screens first, then pulled the color, type and spacing standards out of them. The team pulled our frames straight into code through the Figma MCP and was building new screens from the library every day.
Next · PolarMaking long-running work legible.Building something like this?
Start a project