Applied AI At Work / Role Playbooks
Designing AI Experiences
Patterns designers can use to make AI features understandable, interruptible, and trustworthy.
Patterns designers can use to make AI features understandable, interruptible, and trustworthy.
Use the brief to sharpen a real ai upskill conversation: what is the decision, what evidence matters, and what should remain human-led?
Capture one design rule you would reuse when reviewing an AI workload, assistant, or operating model.
Executive note
Core Idea
AI experiences must make the system's role clear. Is it drafting, recommending, summarizing, deciding, routing, or acting?
Users should not have to guess how much authority the AI has.
Section 2 of 6
Make Uncertainty Visible
Good AI UX does not pretend that every answer is equally reliable. It shows enough context for the user to judge the output.
Useful signals include:
The trick is to show uncertainty without overwhelming the workflow.
- Sources.
- Assumptions.
- Confidence cues.
- Known limits.
- Missing information.
- Actions the system can and cannot take.
Section 3 of 6
Give Users Control
Users need ways to interrupt, correct, retry, inspect, approve, reject, and recover.
This is especially important when agents can use tools. A user may be comfortable with a draft, but not with automatic publishing or record updates.
Section 4 of 6
Design The Handoff
AI often creates a handoff moment. The user moves from generated output to human judgment.
Design that moment carefully:
- What changed?
- What should the user review?
- What is uncertain?
- What action is recommended?
- What happens after approval?
Section 5 of 6
Recommended Practices
- Label AI-generated content clearly.
- Separate draft, recommendation, and action states.
- Avoid hiding sources behind decorative UI.
- Make corrections easy and valuable to the system.
- Design for failure, not only impressive success.
Section 6 of 6
Remember This
Trust does not come from animation alone. It comes from clarity, recoverability, permission boundaries, useful defaults, and honest feedback when the system does not know enough.