Agentic AI Action Patterns
AI Agent Action Confirmation
Confirm consequential agent actions before they run.
Production readyBlockingFriction: mediumArticle 50(1)Article 13
ModalInline confirm card in chat
Problem
Confirm consequential agent actions before they run. This pattern helps address:
- irreversible or costly action
- external side effects
When to use
- Before an agent performs an action with real-world impact
When not to use
- Idempotent, easily reversible actions
Do
- Show what will happen, on whose behalf, and with what data
Don't
- Auto-execute risky actions after passive suggestion
Example copy
short
Confirm: send this email?
contextual
The AI assistant is about to send this email on your behalf. Review and confirm.
detailed
The AI assistant is about to send this email to jane@example.com on your behalf. Review the content and confirm — this action cannot be undone from within the chat.
Implementation checklist
- Explicit user confirmation before execution
- Log confirmed action for audit
- Provide 'cancel' with equal weight
Accessibility notes
- Focus trap in modal
- Primary action clearly labelled
Related patterns
This tool provides product and UX guidance for AI disclosure patterns. It does not provide legal advice and does not determine legal compliance. Please consult qualified legal counsel for binding assessments.