30-minute demo

Goal

Show the agentic loop quickly: evaluate the baseline, assess the model, improve with handoffs, compare results, show hook guardrails, and close with a pull request review.

Have pre-run

A clean baseline branch restored from the lab02-start checkpoint.
A sample Lab 05 baseline run file in evaluation/runs/.
A sample Lab 10 after run file.
A pre-opened issue for the cloud or PR story.
A pre-opened pull request with model changes and at least one Copilot review comment.
Power BI Desktop open to the PBIP if you plan to show model state.
Terminal history with validation commands ready.
Browser tabs for Expected results and the active PR.

Timing table

Time	Segment	Show	Transition sentence
0:00 to 2:00	Setup	Repo map and baseline model	“This repo treats Copilot customization as code, so every instruction, skill, agent, prompt, hook, and check is reviewable.”
2:00 to 6:00	Lab 05 before run	Baseline question-tester run or sample	“Now we need a score, not a vibe, so we ask the model the same business questions every time.”
6:00 to 10:00	Lab 06 assessor	`model-assessor` findings	“The assessor does not fix anything. It gives us a bounded model-improvement plan.”
10:00 to 15:00	Lab 07 handoff	Handoff to `model-improver`	“This is where the workflow becomes agentic: the right agent takes the next step with the right tools.”
15:00 to 20:00	Lab 10 comparison	Before and after score	“The proof is not that files changed. The proof is that the same questions get better answers.”
20:00 to 24:00	Lab 11 hook denial	Denied edit or command	“Instructions are guidance. Hooks are policy.”
24:00 to 29:00	Lab 12 PR review	PR, CI, model-reviewer, Copilot review	“The last mile is a pull request that humans and agents can both review.”
29:00 to 30:00	Close	Next labs	“If you have 60 minutes, add deterministic DAX tests, cloud agent setup, and your own lab extension.”

Segment script

0:00 to 2:00, setup

Open the repo root and point at .github, fabric, evaluation, scripts, and site.

Say: “The baseline model is intentionally not AI-ready. The lab teaches the work needed to make it easier for Copilot and Power BI users to reason over the model.”

Link: Lab 02.

2:00 to 6:00, Lab 05 before run

Run or show a saved baseline question-tester run. If running live, use the same five questions and save the JSON run.

Say: “This is our before score. We keep the questions stable so improvements are measurable.”

Link: Lab 05.

6:00 to 10:00, Lab 06 assessor

Open the model-assessor agent or a saved assessment. Highlight missing names, descriptions, relationships, measures, and AI metadata.

Say: “Notice the assessor is constrained. It inspects and recommends. It does not rewrite the model.”

Link: Lab 06.

10:00 to 15:00, Lab 07 handoff

Show the handoff from assessor to improver. Open changed TMDL files and the report binding work.

Say: “The handoff keeps the finding, scope, and next action together, but you still review before the next agent runs.”

Link: Lab 07.

15:00 to 20:00, Lab 10 comparison

Show Expected results and the before and after score.

Say: “The AI-ready model does not just look cleaner. It gives the assistant better evidence for the same business questions.”

Link: Lab 10.

20:00 to 24:00, Lab 11 hook denial

Trigger a safe denial, for example an attempted edit to data/generated/ or a protected expected-answer read in a non-maintainer session.

Say: “This is the difference between hoping agents behave and enforcing the rules in code.”

Link: Lab 11.

24:00 to 29:00, Lab 12 PR

Open the pre-created pull request. Show CI, the PR template, CODEOWNERS, model-reviewer, and Copilot review comments.

Say: “The repo is built so agents can help before, during, and after the pull request without bypassing review.”

Link: Lab 12.

Fallbacks

If the live agent stalls, show the saved run. If Desktop is slow, skip Desktop and show file diffs plus validation. If GitHub review is delayed, use the pre-recorded PR tab.