AI DEMOS · USE CASE 02 · STATUS REPORTING
The weekly status,
written from evidence.
A demo of status reporting done differently: one messy sprint summary given to Claude, ChatGPT and Gemini at the same time, each drafting the status in the three-line evidence format. Every draft names the model that wrote it, and the page closes with the joint discussion.
- Fictional demo — illustrative answers
- Not a real project
- Model versions labelled
THE SETUP
What was asked,
and how.
The input is the kind of notes a delivery lead actually has on a Friday: standup scraps, a slipped integration, a vendor gone quiet. One prompt, sent to all three assistants the same day.
“Here is this week’s raw delivery notes. Write the weekly status in three lines only: (1) what we saw working this week, (2) the date we are least sure about and why, (3) the decision we are waiting on and from whom. No colours, no percentages — evidence only.”
THE ANSWERS
Three draft statuses.
Each draft below is illustrative demo content written to show the format. The version badge makes it auditable — rerun the prompt against the same model and compare.
ANTHROPIC
Claude
claude-sonnet-5-5- Saw working: payment-matching service deployed to staging; 1,200 test transactions cleared with three defects raised and owned.
- Least sure: 14 Nov for the fraud-rules integration — the vendor has not confirmed the test-window slot.
- Waiting on: sign-off on the chargeback scope cut, from Petra (product).
Sharp on uncertainty; says “not confirmed” instead of guessing a new date.
OPENAI
ChatGPT
gpt-6-sol- Saw working: reconciliation defect rate down to 0.4% after the fix; onboarding flow passed review with two minor notes.
- Least sure: the 14 Nov fraud-rules date — flags it as the only date that moved twice.
- Waiting on: a written decision on scope by Wednesday, otherwise the regression pack cannot be frozen.
Strongest on the ask: turns the decision into a dated, owned request.
Gemini
gemini-3.8-pro- Saw working: lists the demo, not the claims — staging deploy ran end-to-end in front of ops on Thursday.
- Least sure: 14 Nov; separates the risk (vendor slot) from the comfort (fallback date 21 Nov exists).
- Waiting on: product decision on chargeback scope; names the meeting where it can be closed.
Best “we saw it working” — evidence a reader could check, not adjectives.
JOINT DISCUSSION
What the three said together.
Three drafts, zero colours, zero percentages — and every one of them fit in three lines. The joint discussion is what a team would take from running this weekly.
Where all three agreeThe 14 Nov fraud-rules date is the weak point and the chargeback scope decision is the blocker. No model padded either — the format does not invite it.
Where they splitTone, not facts. Claude writes for a steering meeting, ChatGPT for an owner who must act this week, Gemini for an auditor who asks “how do you know”. Pick per audience — or let a human merge in two minutes.
What a human decidesWhether the two facts are true. The models structure the status; a person still owns the number that goes out and who hears it first.
WHY THIS FORMAT
Three lines that are hard to fake.
“What we saw working” dies after one week of fiction. A named date with a reason survives a steering meeting. A decision with an owner gets made. That is the whole trick — and it is what the demo shows, not tells.
Demo content. The drafts above are fictional illustrations of the format — not real model outputs, and not a real project.
WANT THIS ON YOUR OWN SPRINT?
We will build the same
demo for your case.
Your tooling, your cadence, the same auditable format — usually ready within days of a first call.