Trust Calibration

Design a system that progressively builds appropriate trust through demonstrated competence - showing track records per domain, celebrating milestones, and adjusting oversight based on actual agent performance.

What is Trust Calibration?

Users either over-trust or under-trust AI agents. Over-trust leads to passive reliance on inaccurate outputs where users stop checking and mistakes compound. Under-trust means users micromanage every action, defeating the purpose of delegation. Trust calibration is the design challenge of aligning a user's perception of the agent's reliability with its actual performance over time. Unlike one-time confidence scores, this is a relationship that evolves - the agent earns more or less trust based on its track record with that specific user. The pattern starts agents supervised with high visibility, shows per-domain track records, proactively repairs trust after mistakes, and offers autonomy upgrades only when earned. Trust builds slowly and breaks quickly, and the design must account for this asymmetry.

Problem

Users either over-trust or under-trust AI agents. Over-trust leads to missed errors; under-trust leads to micromanagement. Trust calibration aligns user perception of agent reliability with actual performance, but it evolves over time per domain.

Solution

Build appropriate trust through demonstrated competence: start supervised, show per-domain track records, celebrate milestones, proactively repair trust after errors, and only offer autonomy upgrades when performance warrants it.

Real-World Trust Calibration Examples

Implementation

When to use Trust Calibration, and when it backfires

Use it when

The agent acts over time and across domains, so a single static confidence score would misrepresent it. Trust needs to evolve with the track record.
Mis-set trust is costly: over-trust lets errors compound unnoticed, under-trust makes users micromanage and abandon the agent.
Reliability genuinely varies by domain, so a blanket 'trust the AI' is wrong and a per-domain record actually means something.

Don't, or minimize, when

The interaction is one-shot or stateless. There's no relationship to calibrate; a per-output confidence score is the right tool, not a track record.
You don't actually measure outcomes. A trust score with no performance data behind it is theater, the same calibration lie as a fabricated confidence number.
Reliability is uniformly high or the stakes are trivial. Elaborate trust-building UI is just friction.

The trap

The vanity trust score: a 'trust level' that climbs with usage or time rather than with measured accuracy. It manufactures trust the agent hasn't earned, encourages the exact over-trust the pattern exists to prevent, and collapses the first time a 'highly trusted' agent makes a visible mistake. Trust must track competence, not engagement.

Take it into your own product

1
Start supervised, earn autonomy.
Default a new agent to high visibility and human-in-the-loop, then widen its latitude only when its track record warrants it. Granting autonomy on day one is borrowing trust the agent hasn't earned, and the bill comes due on the first unattended mistake.
2
Show the track record, per domain.
'Trustworthy' is not global. An agent excellent at scheduling may be unreliable at spending. Show competence per domain so users calibrate where it actually matters, instead of collapsing everything into one misleading score.
3
Tie the trust signal to performance, not usage.
A trust level that rises with time-spent or clicks is a vanity metric. It has to move with measured accuracy and outcomes, or it's the same lie as a fabricated confidence number, and it quietly trains users to over-trust.
4
Repair trust proactively after a mistake.
Trust builds slowly and breaks fast. After an error, surface what happened, what changed, and dial oversight back up yourself. Don't wait for the user to lose faith in silence and walk away, you rarely get told why they left.
5
Treat under-trust as a failure too.
If a user is double-checking every action the agent reliably gets right, calibration has failed on the other side: the agent is being micromanaged into uselessness. Surface the track record to earn back appropriate delegation, not only to warn.

Apply with Claude Code

Add Trust Calibration to your product

Copy the prompt below into Claude Code or Cursor in your repo. It encodes the four moves on the left and asks Claude to find your AI decision surfaces and update them. Claude reports what it changed and asks before adding dependencies.

30-second check

Check if your product already has this pattern

Upload a screenshot. We'll tell you which of the 36 patterns your AI interface uses and where the gaps are.

Audit My Design

Practice in Courses

Claude Code

Claude Code Course for Designers

23 lessons — free course

Cursor

Cursor Course for Designers

12 lessons — free course

GitHub Copilot