Pattern 08 · Dashboard with AI insights · Interaction pattern
Review a colleague's work checkpoint
Practice view. What to build, the words on screen and how to measure it. Switch to Research for why it should work, the framework properties and the references.Research view. The research side is shown below the card: why it should work, the properties it targets and the references. Switch to Practice to hide it.
When a junior completes an AI-assisted task, it appears in the senior's queue as a review item. The item carries the original request, what the AI prepared and what the junior changed, with two actions: approve or request changes.
The user stays in chargeThe system shows how it worksCase: CNEXT Mail2Flow (company approval pending)
Done right, done wrong
Schematic screens, not real products. Click a screen to enlarge it.In practice
The July script asked the accountants to find a VAT workbook prepared by a junior, evaluate it and decide. It was the easiest task of the session: all seven read the summary, looked at what the junior had submitted, and approved or edited the draft. One reason it went well is that the screen held one proposed action, where the earlier task had shown two side by side (see One proposed action, one named button).
Two observations sharpen the pattern. One participant pressed Approve and got no confirmation, so she did not know whether the response had gone out; the checkpoint needs a closing line, sent or filed, under whose name. And one senior followed the AI draft although she said afterwards she would have written something different, because the text was already there. A review card should make editing as easy as approving, or the checkpoint becomes a rubber stamp.
Words on screen
Obligations to keep in mind (AI Act, GDPR)
- Human oversight with enough context to decide; show what the AI did and what the person changed. (O2, IUI 2026, Table 1)
Design considerations, not legal advice or a conformity assessment. What O1 to O5 mean.
Build it
Illustrative signatures, not a published library: the names below are suggestions to adapt to your stack. Thresholds are examples to tune on your own data.
States: pending, approved, changes-requested
Component: <ReviewCard request={e-mail} draft={draft} edits={diff} onApprove={fn} onChanges={fn} />
Measure it
Same idea, other names
Applied by: nobody yet · Replicated by: nobody yet · Used this pattern? Tell us and get listed
Try it
Research side · why it should work
Why it should work
Properties targeted
Across contexts
| Context | Users | State | Result |
|---|---|---|---|
| CNEXT Mail2Flow, Kairos prototype | accountants, desktop | Seen in a case | 7 of 7 completed the review task; one got no confirmation after Approve; one followed the AI draft although she would have written differently |
| Your context | Bring your case |