Designing for human + agentic capabilities

Agentic systems change the UX contract. Users are no longer only clicking through fixed flows — they negotiate goals with software that can take multi-step action.

Useful patterns so far: explicit goal confirmation, inspectable plans, reversible actions, and clear ownership of what the agent may do unsupervised versus what stays human-gated.

The research question underneath is familiar from HAT work: how do we leverage both human judgment and agentic speed without collapsing accountability?