Operator study · 4 modules · agenda published · pilot cohorts forming

AI in the SDLC: from ticket to production

Teach a team where AI helps, where evidence is mandatory, and which controls cannot be delegated to a plausible-looking diff.

Open the 18-page control map Ask about a pilot cohort

Who it is for

A working team, not a prompt hobbyist

Engineering managers introducing coding agents
Security and platform teams defining acceptable use
Developers who review AI-generated code
Product leaders measuring delivery outcomes

Credential language: AI Adoption Trends can issue a completion record for a facilitated cohort. This is not an accredited certification and is not presented as one.

Launch status: The syllabus, exercises, capstone, and completion standard are published. Facilitation notes are being refined with early cohorts, so fit and format are confirmed before a session is proposed.

Frame the work before the model sees it

Turn a feature request into a clearly limited AI-assisted task with an owner, acceptance tests, and a data rule.

Exercise: Write a task card for one backlog item and mark what may and may not enter a model context.

Control context, dependencies, and generated code

Separate useful repository context from secrets, customer data, and unreviewed third-party code.

Exercise: Build a context manifest and a dependency-review checklist for the task card.

Test behavior, not fluency

Use deterministic tests, adversarial cases, and human review instead of treating a plausible diff as evidence.

Exercise: Add one unit test, one misuse case, and one rollback trigger before accepting an AI-generated change.

Ship with source records and a rollback path

Record model/tool use, reviewer responsibility, release evidence, and post-release monitoring.

Exercise: Complete a release record and run a 20-minute rollback scenario exercise.

Assessment design

Evidence from one real change

The final project is a reviewed change packet: a clearly limited task, a list of the context supplied to the AI, test evidence, human approval, dependency record, release note, monitoring trigger, and rollback path. A polished prompt without those records does not pass.

Primary sources and further study

Pilot cohort

Run it against your repository and release process.

Send the team size, stack, and release constraint. Gary will reply with fit and possible next steps. This does not reserve a date or collect payment.