Overview
How might we design interruptibility so people can trust and act on AI output?
When this pattern fits
- Essential for voice assistants, long-running AI agents, and streaming responses where users need a single, reliable gesture to pause, cancel, or redirect the system mid-execution.
When to skip or lighten it
- Operations that are unsafe to interrupt mid-write without rollback (disable interrupt or make it transactional).
- Instant actions that finish before a cancel target can be hit.
- Read-only playback where pause is enough and cancel is meaningless.
States
State model coming soon
Key UX elements
Key UX elements coming soon
Anti-patterns to avoid
No stop control during long generations.
Cancel that leaves partial side effects applied.
Tiny hit targets during voice mode.
Requiring a deep menu to stop an agent.
How products use it
| Product | Implementation |
|---|---|
| ChatGPT Voice | Tap/interrupt to stop speaking or listening. |
| Siri | Tap or speak cancel to halt the assistant. |
| Alexa | Stop commands during skills and speech. |
| Cursor Agent | Stop/pause controls during agent execution. |
Implementation
Copy this prompt to generate a production-ready implementation in Cursor, Claude Code, Lovable, or any AI coding agent.
Generate a production-ready implementation of the "Interruptibility" AI interface design pattern.
Pattern Definition:Frequently asked questions
What makes interruptibility feel trustworthy?
An always-visible stop, immediate halt of new side effects, and a clear leftover state (partial text kept or discarded).
Interrupt vs undo?
Interrupt stops in-progress work. Undo reverses completed work. Agents often need both.
Where should the control live?
Replace send with stop during generation, and keep a floating stop in voice/agent modes.
What about keyboard and accessibility?
Support Escape or an equivalent, and announce state changes to assistive tech.