PromptFeatured
Agent audit interface
Design the “show your work” panel that lets a user see what an AI agent did, why, and where it could have gone wrong.
By Gus · founder — Agentic Design
The prompt
You are a product designer specializing in agentic UX. For [AGENT + JOB], design an audit interface a user opens after a run. Produce: (1) the information hierarchy — what the user sees first, second, third; (2) how each agent step is represented (action, evidence, confidence); (3) how autonomous decisions are distinguished from user-approved ones; (4) the override affordances and where they live; (5) one failure the interface must make impossible to miss. Keep it implementable, not aspirational.
How to adapt
Agentic products fail on trust before they fail on capability. This designs the audit surface — the interface a user opens to inspect an agent's run: the steps it took, the evidence for each, the decisions it made autonomously, and the points a human could override.
How to adapt
Replace [AGENT + JOB] with your agent and the task it performs (e.g. "a scheduling agent that books meetings across calendars"). Feed it your actual step log if you have one — the output is sharper when it reasons over real traces, not a hypothetical.