Slide 1
Slide 1 text
Speed Bumps, Not
Roadblocks:
Designing AI
Trust
Slide 2
Slide 2 text
Hello
DariaTarawneh
Head of Enterprise Design Miro
Slide 3
Slide 3 text
No content
Slide 4
Slide 4 text
No content
Slide 5
Slide 5 text
No content
Slide 6
Slide 6 text
No content
Slide 7
Slide 7 text
No content
Slide 8
Slide 8 text
Frictionless + high stakes =
disaster.
Slide 9
Slide 9 text
No preview of scope (247 files would be affected)
No risk assessment ( PRODUCTION ENVIRONMENT)
No pause/stop button during execution
No real-time activity log ("Currently deleting payment.js...")
No confidence indicator ("I'm 40% sure about this approach")
No "Wait, that seems extreme" warning
And not compliant with rules and regulations.
The design failures:
Slide 10
Slide 10 text
What Marcus needed wasn't ease.
What Marcus needed was speed bumps.
Slide 11
Slide 11 text
ACT 1: The reality of AI adoption
Slide 12
Slide 12 text
78% of companies in the US
have deployed AI in their
organisation.
Slide 13
Slide 13 text
what is the percentage
of AI activation in
companies in Japan?
30%
90%
50%
70%
Slide 14
Slide 14 text
what is the percentage
of AI activation in
companies in Japan?
60%
90%
50%
70%
30%
Slide 15
Slide 15 text
84% of companies world wide say security
and compliance are their #1 buying criteria.
Slide 16
Slide 16 text
Only 9.6% of Japanese AI users would
feel comfortable handing over full
control to AI for business, but 55.2%
would be open to using AI with review or
oversight mechanisms in place
Slide 17
Slide 17 text
"Show us exactly what it's doing"
"Let us say 'no' to specific things"
"Our rules, not yours"
"Prove to us people are using
this safely"
The shopping list
Slide 18
Slide 18 text
How Miro does it
Slide 19
Slide 19 text
No content
Slide 20
Slide 20 text
No content
Slide 21
Slide 21 text
No content
Slide 22
Slide 22 text
No content
Slide 23
Slide 23 text
No content
Slide 24
Slide 24 text
In general companies are paying 15-30%
premiums for security and control features.
Slide 25
Slide 25 text
ACT 2: The Trust framework
Slide 26
Slide 26 text
VISIBILITY
The Trust Design
Framework
CONTROL
PREDICTABILITY
ACCOUNTABILITY
RECOVERY
Can I see what's happening?
Can I stop it if it goes wrong?
Does it behave consistently?
Can I trace what happened?
Can I fix it when it breaks?
Slide 27
Slide 27 text
Visibility: No preview of
what AI would do
Control: No scope
limitations
Predictability: No
indication of how "fix
bugs" would be interpreted
Enter task: _______________
[Submit]
Slide 28
Slide 28 text
Visibility: Full scope
preview
Control: Multiple decision
points
Predictability: Shows AI's
interpretation
🤖 AI Task Analysis Your request: "Fix all files in production" Scope detected:
247 files in production environment
⚠️ Includes critical systems: • Payment processing • User authentication • Core
API AI's proposed approach: "Delete files containing bugs"
⚠️ Risk Assessment: CRITICAL
Confidence in this approach: 42% (LOW)
Alternative approaches available: - Flag files for human review - Create fixes
without deletion - Generate detailed bug report [ Cancel ] [ Choose different
approach ] [ Set safety limits ] [ Proceed with extreme caution ]
Slide 29
Slide 29 text
ACT 3: The Designers Toolkit
Slide 30
Slide 30 text
Visual language that changes based on how confident
AI is.
Why it matters:
When all outputs look the same, users can't
distinguish reliable from risky.
Apply to:
Background tints
Border styles (solid
→ dashed
→ dotted)
Typography weight
Icon treatment (filled
→ outlined)
Button states
Confidence Gradient
Making AI's thought process visible and traceable.
Why it matters:
Black box decisions break trust. Transparent
reasoning builds it
Always show source count
Make sources clickable
Log the AI's reasoning steps
Present in digestible chunks
Source Chain
Slide 31
Slide 31 text
Specific moments where humans can
review, modify, or stop AI actions.
Why it matters:
Control without intervention points is
an illusion.
The Intervention
Point AI suggests action "Delete 8 duplicate files" Affected files: contacts.csv, data.xlsx...
[ Review files first ] [ Proceed ]
APPROVAL (High-risk)
⛔ AI needs approval "Send email to 2,847 customers" This action cannot be undone.
You must review before proceeding: ☐ I've reviewed the email content ☐ I've verified the
recipient list ☐ I understand this cannot be undone [ Cancel ] [ Approve and send ]
EMERGENCY STOP (Always visible)
🚨 [STOP ALL AI ACTIVITY]
Slide 32
Slide 32 text
A record of what AI did, why it did it,
and what happened.
Why it matters:
Accountability requires traceability.
You can't fix what you can't see.
Scannability: Visual hierarchy, not
walls of text
Actionability: Every entry has action
buttons
Searchability: Filter by action type,
date, confidence
The Audit Trail
Key elements:
User or AI
: Action taken
Timestamp
Reasoning
Confidence score
What actually happened
Available actions
Who:
What
When:
Why:
How certain:
Result:
Recovery:
Slide 33
Slide 33 text
We're not just "UX designers" anymore.
We're trust architects.
Slide 34
Slide 34 text
Thank you
Daria Tarawneh
Connect with me