Build and review code with AI
You’ll work on a prepared task API with an incomplete request and an AI-generated change that needs review.
The aim is to explain whether that change should ship, using the requirements, diff and test results.
- Format
- 3-hour hands-on workshop
- Bring
- Comfort with code, Git & basic tests
Work through one code change
You write acceptance criteria and supply the context the assistant needs before generating or revising code. Then you inspect the diff, test expected behaviour and failure cases, and check the data and dependencies involved. Use what you find to decide whether to ship, revise or reject the change.
You leave with a reviewed change and a reusable review and testing checklist. The exercise uses synthetic code; you won’t need to bring a private repository.
Try the review checklistExamples from my projects
I bring examples from Fluent, where captions and translated audio have to reach a live audience; Mirror, where camera processing stays in the browser; and Passport, where an offline guide still depends on online generation.
We discuss those decisions; building all three products is outside the afternoon’s scope.
Prepare for the session
Before the session, we agree one supported stack, an approved AI tool and the learning objective. We check accounts, licences, network access and a fallback environment so participants can spend the session on the exercise.
See a sample three-hour agenda
| Activity | Minutes | Activity and output |
|---|---|---|
| Baseline and setup | 15 | Confirm the repository runs and define the small task. |
| Specification and context | 20 | Write acceptance criteria, constraints, and an implementation plan. |
| Implementation and review | 35 | Generate a change, inspect the diff, and identify assumptions that need checking. |
| Break | 10 | A break before tests and failure cases. |
| Tests and failure cases | 35 | Check expected behaviour, boundaries, and cases the generated answer missed. |
| Data and dependency boundaries | 25 | Review the data sent to the assistant, new dependencies, and assumptions that need checking. |
| Release decision | 25 | Use the review and test results to explain whether to ship, revise or reject the change. |
| Debrief | 15 | Build a reusable checklist and choose one practice to try at work. |
Prerequisites & scope
Participants should be comfortable reading code, using Git, and working with basic tests. This is AI-assisted software practice, not an introduction to programming or a foundation-model training course.
A longer session can add another exercise or deeper deployment work. The workshop does not certify software as secure or production-ready. Materials are selected with their reuse rights in mind.
Plan a developer workshop
Tell me your stack, how you’re using AI today, and what you’d like your developers to do better.