Hands-on labs · EN / FR · Windows + PowerShell + VS Code
Idea to production, with an AI team and a human in charge¶
You will use the ai-team-sdlc GitHub Copilot plugin: an orchestrator and ten specialist agents that design, build, test and deploy an app, with quality gates at every step and one mandatory human sign-off.
What you will build¶
In the recorded run, the agents build Pinch: a small, pretty and fully bilingual (EN/FR) recipe scaler with a cook mode. It has no backend and no secrets, so it is safe to build, review and deploy in a few hours.
- Scale a recipe to any number of servings, with sensible rounding.
- Convert between metric and imperial units.
- Cook mode: large text, tick off steps, build a shopping list.
- Bilingual UI: every string exists in English and French, enforced by a gate.
The app lives in its own repository, ai-sdlc-labs-pinch, with a checkpoint tag at the end of every lab.
The lifecycle at a glance¶
- Kickoff Idea product brief
- Plan Plan design · prototype · spec review
- Build Build build · lint · unit
- Test Test qa · review · security
- Sign-off Sign-off mandatory human approval
- Deploy Deploy pre-deploy · smoke · rollback-ready
- Shipped Production DONE
Two loops you will meet in the labs
Gate fail: when a quality gate fails, the work goes back to the Build phase and the owning agent fixes it. Changes requested: at sign-off, a human can send the work back to Build. Any change after approval resets the sign-off.
Meet the team¶
Eleven ait- agents work as one team. Humans stay in the loop at every phase.
-
Orchestrator
ait-sdlc-orchestratorSplits the work into tasks, dispatches agents, enforces the gates. -
Product Designer
ait-product-designerUser journeys, wireframes, a clickable prototype. -
Product Owner
ait-product-ownerRequirements, acceptance criteria, backlog. -
Architect
ait-architectTechnical spec, data model, decisions (ADRs). -
Backend Dev
ait-backend-devServer logic, APIs, persistence, tests. -
Frontend Dev
ait-frontend-devUI, accessibility, client state, tests. -
QA / Test
ait-qa-testTest plan, acceptance checks, regressions. -
Code Critic
ait-code-reviewerFinds blocking defects before sign-off. -
Security / RAI
ait-security-raiSecrets, dependencies, threats, responsible AI. -
DevOps
ait-devopsCI/CD, release checks, rollback plan. -
Scribe
ait-scribeKeeps the decisions and the change log tidy.
The labs¶
| Lab | Title | Duration | Level |
|---|---|---|---|
| 0 | Prerequisites and setup | 20 min | Beginner |
| 1 | From idea to brief | 25 min | Beginner |
| 2 | Design and prototype | 35 min | Beginner |
| 3 | Requirements and architecture | 35 min | Intermediate |
| 4 | Build in slices | 60 min | Intermediate |
| 5 | QA, critic review and security | 45 min | Intermediate |
| 6 | Human sign-off and changes requested | 30 min | Intermediate |
| 7 | Deploy to production | 30 min | Intermediate |
| 8 | Resume, recover and tear down | 25 min | Intermediate |
Two ways to learn¶
Follow the recorded run¶
Read each lab, then jump to its checkpoint: a Git tag in the Pinch repository that holds the exact result of the recorded run. Cheap, fast and repeatable. Use it to learn how the gates and artifacts look without spending credits.
Bring your own idea¶
Run every lab on your own product idea in your own repository. The agents' output will differ from the recorded run, and that is expected. A full lifecycle run can use about 2,000 credits and 2 to 3 hours.
Mind the budget
Agents run with broad permissions and real model credits. Read the cost note in Lab 0 before you start a long run.