60-minute demo

Goal

Run the same story as the 30-minute demo, then add deterministic DAX tests, the Copilot cloud agent workflow, and a short build-your-own customization exercise.

Have pre-run

Everything from the 30-minute demo.
Power BI Desktop open with the AI-ready checkpoint for Lab 08.
A completed Invoke-DaxQuestionTests.ps1 output.
A cloud-agent issue and branch ready to show.
A small Lab 14 customization idea prepared, such as a new prompt or skill reference.

Timing table

Time	Segment	Show	Transition sentence
0:00 to 4:00	Setup	Repo map, baseline, customizations	“The repo is the product: model code, agent instructions, guardrails, checks, and docs live together.”
4:00 to 10:00	Lab 05 before run	Baseline question-tester	“First we measure the assistant on business questions before we improve the model.”
10:00 to 16:00	Lab 06 assessor	Assessment and skill loading	“Skills give the agent domain knowledge without stuffing every rule into the main prompt.”
16:00 to 25:00	Lab 07 handoff	Improve names, descriptions, relationships	“Handoffs let the assessing agent pass focused work to the improving agent.”
25:00 to 34:00	Lab 08 DAX tests	Measures and deterministic tests	“Now we separate language quality from calculation correctness.”
34:00 to 41:00	Lab 10 comparison	Before and after results	“The same five questions become our regression test for AI readiness.”
41:00 to 46:00	Lab 11 hooks	Denial, ask, post-edit lint	“Hooks give us executable guardrails around agent behavior.”
46:00 to 53:00	Lab 12 PR	CI, reviewer, Copilot review	“The workflow ends where production changes should end: in a reviewed pull request.”
53:00 to 57:00	Lab 13 cloud agent	Cloud setup and issue	“When work is scoped and scripted, you can hand it to a cloud agent without losing control.”
57:00 to 60:00	Lab 14 in five minutes	Add one customization	“The same pattern works for your domain: write the rule, wire the agent, test the outcome.”

Segment notes

Setup

Say: “There is no hidden service dependency for the lab. The dataset, oracle, grader, model files, site, and checks are all in this repo.”

Lab 05 before run

Show a baseline run and the JSON shape. Keep the focus on repeatability.

Link: Lab 05.

Lab 06 assessor

Show the skill catalog and the model-assessor agent. Point out that the assessor can read model files and rules but should not make edits.

Link: Lab 06.

Lab 07 handoff

Show the handoff card or prompt from assessor to improver. Then show the actual changed TMDL names and descriptions.

Link: Lab 07.

Lab 08 DAX tests

Run or show:

pwsh ./scripts/Invoke-DaxQuestionTests.ps1

Say: “The natural-language evaluator tells us whether the assistant can answer. The DAX tests tell us whether the model can calculate.”

Link: Lab 08.

Lab 10 comparison

Open Expected results and the comparison output.

Say: “The expected answers came from DuckDB, not the semantic model, so a broken model cannot grade itself.”

Link: Lab 10.

Lab 11 hooks

Show a denied protected edit, an ask decision for destructive cleanup, and a post-edit lint message if you have one.

Link: Lab 11.

Lab 12 PR

Open the PR. Show validation, template fields, CODEOWNERS, model-reviewer notes, and Copilot review.

Link: Lab 12.

Lab 13 cloud agent

Show .github/workflows/copilot-setup-steps.yml, the cloud issue, and the cloud agent branch or PR.

Say: “Cloud agent works best when the repo already has setup steps, tests, instructions, and tight issue scope.”

Link: Lab 13.

Lab 14 in five minutes

Pick one small extension: a prompt, a skill reference, or a rule update. Do not build a full new lab live.

Say: “Build-your-own does not start with a giant agent. It starts with one repeated task and the smallest customization that makes it reliable.”

Link: Lab 14.