We design the moment a person signs off Screens where people and AI agents share the work
When an agent does part of the job, the screen has to show what it did, why, and how sure it is, and leave the decision with a person.
A refund request, order 4471
The agent suggestsApprove a refund of $86.40
WhyThe order arrived nine days late. Your policy allows a refund after seven.
How sureHigh
ApproveChange the amountSend to a manager
Nothing is sent until you approve, and you can undo it from the order's history.
Shows what the agent didShows whyShows how sure it isThe person decidesIt can be undone
What the lab does For products your people use every day
Design systems your agents can build withWe build your design system and serve it to coding agents, so new screens follow your brand from the first draft.
Screens for work that agents shareWhere the person signs off, how the agent shows how sure it is, and how anyone undoes its work.
Fix the product you already haveWe study it with real users and analytics, then redesign it in releases, with no big-bang relaunch.
Design new products from researchInterviews and tests with real users first, then prototypes, then the design system your engineers build from.
Your design system, served to your coding agents Agents stop guessing at colors, spacing, and components
Coding agents now write a lot of front-end code. Without your design system, they guess. We build it and serve it as an MCP server, so every agent asks it before it draws a screen.
We run our own brand this way. Damco's designers, marketers, and agents all read one brand MCP server before they make anything.
# Before it builds a screen, the coding agent asks → design_system.get_component("approval card") ← Approval card use when an agent proposes something a person approves must show what it did, why, how sure, how to undo primary "Approve", one per screen agent copper square person ink circle radius 14 # Your designers change the system once, # and every agent builds with the change.
Six rules for screens with agents in them We design every agent product to meet all six
- Show when the agent acts, and whyThe person always knows which work the agent did.
- Show how sure the agent isConfidence is on the screen, not hidden.
- The person decidesThere's always an override and a way out.
- Every output can be traced and undoneEach change keeps its history, like code does.
- The work survives a failureWhen the agent fails, the person can still finish the task.
- Every decision can be explainedIn words the person using the screen understands.
Tell us about your product
Tell us who uses your product and what slows them down.