AI Product Design & Agent Interface Trust Checklist

1. Intent Is Visible

• The interface repeats the user's request in plain language before running important work.
• Ambiguous instructions trigger a clarification step instead of a guess.
• The user can edit the interpreted task without starting over.
• The agent shows which files, tools, or context it plans to use.

2. Progress Feels Inspectable

• Long-running tasks show status, current step, and expected next step.
• The user can distinguish thinking, fetching, editing, waiting, and completed states.
• Progress messages are specific enough to reduce anxiety, not decorative loading text.
• Partial results are visible when useful.

3. Control Stays With The User

• The user can pause, cancel, revise, or approve agent actions.
• High-risk actions require confirmation before execution.
• The interface separates suggestions from completed changes.
• Undo or rollback is available for meaningful changes.

4. Recovery Is Designed

• Error states explain what failed and what the user can do next.
• The agent keeps useful context after a failed attempt.
• The user can rerun with adjusted instructions.
• Empty states include example tasks that match real user intent.

5. The System Has A Trust Boundary

Trustworthy agent design does not make the product look magically capable. It shows limits, asks for permission at the right moments, and makes the automation legible enough that users can keep agency.

• Read the broader AI product design checklist
• See the AI product design case study for a voice agent