Product & Strategy
Designing User Experiences for AI Agents
Make goals, progress, evidence, approvals, uncertainty, and control understandable to users.
Start with a clear contract
Explain what the agent can do, what information it needs, and what will require approval. Users should not have to infer the system's authority from a chat box.
Show meaningful progress
Present milestones, current activity, sources, and pending decisions rather than streaming internal reasoning. Let users inspect or correct the working state when it affects the result.
Keep control close
Cancellation, editing, retrying, and escalation should be easy to find. Approval screens should describe the specific action and consequence, not merely ask whether to continue.