AI design critique: reviews that push back
Tipalti's own user personas critique a screen and argue with each other about it, before the real design review does. Designed and built solo on the Claude API.
- Company
- Tipalti
- My role
- Product designer
- Built with
- Claude API
- Timeline
- July – October 2026
One screen. Very different readers.
A CFO, an AP specialist, a monthly approver and an external vendor all use the same Tipalti screens, with very different goals. Design reviews catch usability issues but miss what matters to each of them.
And more non-designers now build UI without any design review.
What I did
Solo · concept to working toolTurned our persona research into AI reviewers, designed how they rank, debate and reach a verdict, and built the tool. Designers use it as a pre-review check. Non-designers building UI use it to get critique on their own.
personas from Tipalti research
panel, usability, accessibility
critical to what works
in design reviews at Tipalti
A panel that disagrees, then a verdict
- 01
Pick the panel
Personas from 4 user groups
- 02
Add the screen
Screenshot, Figma frame or description
- 03
Read the critique
Findings ranked by severity
- 04
Argue back
Pin a comment on any finding
- 05
Get the verdict
A priority list with fix costs
Push back on any finding
Pin a comment and argue back. The persona concedes when you're right.
What makes it useful
Disagreement is the feature
Conflicting goals surface real trade-offs.
Honest severity
If everything is critical, nothing is.
The verdict waits for you
Your arguments shape the final list.
Plain English, by rule
Written for non-designers and non-native speakers.
A side project that took off
I built this for myself, out of curiosity: to see what my idea could become with AI's help. When I showed it to colleagues, it got far more interest than I expected.