About
Meaning is the artifact.
The screens your customers use are projections of something underneath: the objects your product is made of, what they’re allowed to do, and who can do it. We work on that layer, and on how teams work there with AI. Three ideas shape the work.
Read the system, not the surface.
Screen readers have worked without seeing the screen for twenty years. Test tools find elements by role and name, because selectors pinned to position don’t survive a refactor. Both arrived at the same answer: role, name, state and relationships hold up. Appearance doesn’t.
Agents are the newest reader in that family. Hand one a screenshot and it has to reconstruct, as a guess, what the software already knew in explicit form. So we start from the record the software keeps of itself (the DOM, the tokens, the accessibility metadata, the contracts), and every finding points back to the artifact that produced it. It’s the idea behind Stage1 and every Drift Audit.
Give meaning a home.
Design systems standardize the parts. They rarely standardize the meaning. Search one for “User” and you’ll find Avatar, Badge and Card, the pieces used to draw a person, and no definition of what a user is. So every surface decides for itself. The profile, the directory, the picker, the mention and the admin screen each carry their own idea of a user, and the product ships a dozen definitions that drift apart.
We write the object down once, with its traits, relationships, owner and behavior, and treat every screen as a
projection of it. When the object changes, every surface moves with it. Agents need this more than people do. A person
sees a grey button and knows it’s disabled; an agent sees opacity: 0.5. No amount of reasoning downstream
makes up for meaning nobody wrote down upstream. OODS Foundry is how we build screens this way.
Give the work a method.
The models are good enough. What most teams don’t have is a way to work with them together: what shared context looks like when five people use AI on one project, how a decision made in one session reaches the next person in the morning, how the work survives someone closing a tab. Getting everyone onto the tools is adoption, not method. Model capability doesn’t become team capability on its own.
We run work as missions with success criteria, record decisions as they’re made, and hand the context forward, so findings survive handoffs and the work compounds instead of starting over. AI does the scaffolding and the synthesis. People keep the framing and the judgment. We run our own projects this way on CMOS, and it’s what our workshops teach.
The craft didn’t change. The reader did.
What makes design valuable is what it always was: research, framing the problem, understanding who the user is and what counts as a win, and judging the tension between what users need and what the business needs. It never lived in a tool. What changed is who the work has to reach. Some of the next readers are agents, and they read structure, not pictures. We help teams make sure the intent survives the trip.
Start a conversation