
01
HOME02
ABOUT03
BLOGS04
ACTIVITIESMedium

A product I’m building: the control center for your product. It owns your numbers and helps you improve and run your product, on its own or with your team.
Scout AI, the product’s AI teammate, researches, designs and ships changes alongside the team on a shared canvas. But reviewing that work was easy to miss and hard to trust.
How might we make reviewing an AI teammate’s work feel like reviewing a colleague’s: you see what changed, know what it needs from you, and trust that your answer counts?
A review that lives on the work. It shows what changed, reaches the right person wherever they are, and records every yes.
satisfaction with reviewing Scout’s work.
to review, from about 2 min to 40 s.
reviews, down from 3 of 5 testers.
Tentative: early signals from 5 moderated sessions
In moderated testing with 5 early users, we found people struggling to review Scout AI’s work.
What did Scout do?What changed here?Did the owner see it?Will it stop and wait?Can I answer where I am?Who said yes?Reviewing Scout AI’s work broke down. People couldn’t see what needed review, missed when they were asked, and couldn’t trust that their answer counted.
We explored multiple ways to ask for a review. Users preferred seeing what was in review right on the work, while they kept working or made changes.

Explored before it



| Why the final won | Chat ask | Modal | Review inbox | On the work |
|---|---|---|---|---|
| Hard to miss | ✕ | ✓ | ✕ | ✓ |
| Shows the actual work | ✕ | ✕ | ✕ | ✓ |
| Goes to the person who owns it | ✕ | ✕ | ✓ | ✓ |
| Doesn’t stop other work | ✓ | ✕ | ✓ | ✓ |
It lives on the work, and follows people wherever they are.




Always on the work. If you’re on the canvas, nothing else. Outside the app only when you’re away, on one channel, plus one weekly email.





Review the Initial diagnosis for mobile checkout: 3 findings, 2 observed, 1 hypothesis.
Review in appApprove



main from scout/banner

Real reviews rarely go to plan. We designed for when people ask for changes, ask for something Scout can’t do, disagree with each other, aren’t sure who should review, or never reply. In every case the work keeps moving and nobody’s answer gets lost.









Satisfaction with reviewing Scout’s work
Faster to review, from about 2 min to 40 s
Missed reviews, down from 3 of 5 testers
Reviews had to reach people who weren’t in the app, without flooding them.
Canvas, chat, control center, Slack, email and GitHub all had to agree on a single answer.
How Scout asks decides whether a team trusts it.
One clear ask to the right person beats a gate on every step.
People approve faster when the ask sits on the thing they’re approving.
The review lives on the block, with a plain question and one tap answers.
Nothing reaches users without a named person’s yes.
Real components, tested with early users and raised as pull requests.
Track time to approve and missed reviews from the first session, not after.
Decide where an ask goes before building each channel.
Medium
Copyright ©2026 - Designed and developed by Pranjal Choudhary