Pattern Summary
Ask + Confirm is a core agentic interaction where the system proactively gathers an input from the user and explicitly confirms it before proceeding. This pattern reinforces trust, reduces risk, and ensures clarity in intent alignment.
This is especially valuable in moments where the user’s input may trigger irreversible actions, have high business impact, or where model interpretation could be ambiguous.
When to Use It
Use Ask + Confirm when:
- The user input carries operational, financial, or compliance risk
- The AI-generated input or inferred intent might be ambiguous
- You want to maintain user trust by showing caution or respect
- The next action is irreversible or sensitive
- You’re designing for new users who may not yet trust the system
Examples include confirming cancellation of a subscription, validating pre-filled information, or asking the user to verify a suggested response before sending.
How It Works
- Ask Phase: The agent initiates a question or presents a suggested action or value
- User Input: The user provides or accepts an input
- Confirm Phase: The agent summarizes and requests explicit confirmation (e.g., Yes/No or Accept/Revise)
- Execution: Upon confirmation, the system proceeds; otherwise, it branches or halts
This can happen in a single interaction or span multiple steps/sessions.
Fit Assessment
Use this pattern if:
- The user or system cannot act safely on first-pass input
- There is risk in automatic progression
- Your user experience benefits from reflection or checkpoints
Do not use this pattern if:
- You are optimizing for speed above confidence
- Confirmation fatigue is a risk (e.g., users confirm every minor action)
- The input is trivial, easily reversible, or has no material consequence
Acceptable Dependencies
✅ Clear input and confirmation UI elements (buttons, forms, dialogs)
✅ Ability to store intermediate state across steps or sessions
✅ Support for fallback if user revises input
✅ Optional: explanation engine to help user understand why confirmation is needed
Unacceptable Dependencies
❌ Model-only confirmation with no user validation path
❌ Implicit confirmation inferred from lack of user action
❌ Confirmation on every low-stakes task (leads to friction)
Implementation Starter Guide
- Create clear labels: “Are you sure?” or “Does this look correct?”
- Show the input or summary being confirmed, not just a generic message
- Include options for Accept, Edit, or Cancel
- Ensure that confirmations are logged or observable for debugging/auditing
- Design for fallback: if confirmation is denied, what’s the next step?
Example: Lead Qualification Assistant
The assistant suggests a lead priority score based on activity history. Before updating the CRM, it shows:
"This lead has been marked as High Priority based on recent engagement. Would you like to confirm or adjust this classification?"
[Confirm] [Edit] [Dismiss]
Strategic Value
- Increases system transparency
- Protects users and systems from unintended actions
- Reinforces user agency and participation
This pattern builds trust in agentic systems - especially when decisions carry weight. Think of it as the digital version of “double-checking before acting.”
Tags
Pattern Type: Input Validation, Risk Mitigation
Scope: Multi-platform, Cross-role
Recommended UI Modes: Dialog, Modal, Inline Card
