Why every product org ends up with 4 assistants and 3 hubs, and the small pilot I'd run before arguing for a new team.
Most product orgs I’ve worked in have a design system with clear ownership, and a layer above it. The system covers the component layer: buttons, inputs, tables, the tokens that keep them consistent. The layer above it covers how those parts get assembled into a flow, how the navigation decides where a feature lives, and how a header behaves when 3 squads each need it to do something new. That layer of “shared experiences” is where most of the customer’s experience actually happens, and in a squad-based org it belongs to whichever squad touched it last.
Lately the gap has been showing up in a form I’m starting to recognize across the industry: AI features. Every product org right now has the same list under construction. An onboarding helper. A research assistant inside one workflow. A broader assistant someone is scoping. A few “hubs,” each promising to be the one place to see everything about a topic. Our version of that list, when the senior designers who work across the product walked through their in-flight work at our extended design leadership sync, looked the way I’d guess yours does. Every item on it is a reasonable bet on its own, and together they read like 4 products competing to be the same thing, built by teams who mostly haven’t seen each other’s screens.
The design system covers the parts. The layer above it, where 4 features decide what an “assistant” is, tends to belong to nobody.
One of our senior designers called out the cost of that like this: he’s making decisions about entry points, icons, and response patterns with incomplete information, sharing them with whoever will look, and asking when the org plans to put a stake in the ground. A senior IC on the same call made the counterargument, and it’s a fair one: keep building the small pieces, learn from them, and let the strategy grow out of what ships. My push landed somewhere in between. Nobody needs the full taxonomy today, but we do need a V1 of it, a shared vocabulary for what an “assistant” is, what a “skill” inside one is, and what a “hub” is for, so the next 3 teams start from the same nouns.
Our clinical design lead had asked a version of the same question a week earlier: are we designing a product full of hubs, and if so, which one is the front door? Her answer is the one I keep coming back to. The home surface is a navigation hub, the place that surfaces what’s ahead in your day and what needs attention, and the other hubs are where the work gets done. That distinction only holds if someone owns it. Otherwise every hub drifts toward being a home page.
So who owns the in-between? The design system team is the obvious answer and usually the wrong one, because their remit is components, and stretching them across navigation, information architecture, and shared flows dilutes the thing they’re good at. One of my design leaders pitched a different shape, a small team that acts as the other half of the design system, with a mandate to push production changes across workflows rather than publish patterns and hope. I like it. What I’d avoid is arguing for it on the org chart first, because a new box on that chart is the most expensive way to test an idea. The cheaper test is a pilot: 2 or 3 people, a defined set of shared surfaces, a few months, and a rule that they ship rather than document. If the flows get better and the squads stop rebuilding each other’s headers, the case makes itself.
Two things I’d tell y’all if you’re staring at the same gap. First, count your assistants and hubs. If the number surprises you, the layer above your design system is already unowned, and the AI features are just making it visible. Second, write the V1 vocabulary before the strategy. A strategy that arrives after 4 teams have shipped 4 different assistants spends its first year cleaning up.
I build and maintain clinical software, so the stakes for me are a clinician who opens the app and finds 3 places that claim to be the one place. I’d rather they find one, and I’d rather the people building it know each other’s names.