Agents aren't a new audience. They're a new paradigm.

There are two ways to read what’s happening to product design as agents arrive, and you can’t run a roadmap on both. The first says the framework is stable. Agents are simply a new audience, so make your existing patterns legible to them and keep shipping. The second says the framework is discontinuous. New interface categories like Skills, Projects, Artifacts, and Memory don’t appear by extending old ones. They appear when a model crosses a capability threshold, and not before.

The honest resolution is that both are true, at different altitudes. Inside a paradigm, the old principles still govern. Once Artifacts exist as a category, good design is still about labeling, predictability, and the tension between the human reader and the agent one. But between paradigms, no amount of best practice produces the next category. One agent reasoning about another agent’s choreography has no antecedent in human design. That pattern had to be born, not extended.

For a PM the consequence is sharper than either view alone. Before you reach for best practice, ask which paradigm you’re actually in, and whether the model has crossed the threshold that makes the next one real. Polishing a category the model can’t reliably support yet isn’t craft. It’s UX debt.