Internal engineering course

Coding with AI

Learn to build and ship real software with an AI coding agent, from the fundamentals on up.

The normal software lifecycle — plan, build, test, ship — with an AI agent inside every stage and you steering. We call that an AI-augmented SDLC.

Start →
The idea

One repeatable way to take work from a ticket to shipped — with an agent in every stage.

The lifecycle doesn't change shape. What changes is who does each step: the agent does the heavy lifting, and you operate the line — framing the work, supervising, and owning the few calls that stay human.

Set up→ Plan→ Build→ Test→ Review→ Ship An agent works inside every stage. How much it runs alone is a dial — in → on → out of the loop.
The course · thirteen chapters

Foundations

00–06 · the basics, in order
00
How Coding Agents Work
No memory, the context "desk", sub-agents, and where you sit — in → on → out of the loop.
→
↳
Your First Session
One tiny task, keystroke by keystroke — what the screen shows, where you approve, where the change lands. Take it right after 00.
→
01
Working With a Coding Agent
Set the agent up to succeed every task — what it needs to know, and what it needs to run.
→
02
Setting Up the Task
The goal, the boundaries, what "done" means, and the right context.
→
03
Get a Plan Before You Build
Draft a plan with the agent, and approve it before any code.
→
04
Building & Testing
The build-and-test loop, run to green by the agent — one loop for both stacks; test first.
→
05
Backend vs Frontend: What Changes
The little that differs across the stack, then the shared ticket run in two walkthroughs.
→
↳
Backend Walkthrough
The plan's service half, run end to end on .NET — five steps, with the superpowers boxes at each.
→
↳
Frontend Walkthrough
The Angular strip that calls it, proven by a real browser.
→
06
Shipping Your Change
Self-review the diff, open a clean PR, drive CI/CD, deploy, and the production gate.
→

Playbooks

07–12 · run it for real
07
Running the Playbooks
The setup all four playbooks share: which playbook fits the task, point at the repo, set the effort and model, and the three gates you keep.
→
08
Story
The full run, start to finish, on a normal piece of work. The reference the others vary from.
→
↳
Story — On the Loop
The hands-off variant of 08: write the acceptance list, hand the middle to one goal, read the PR at the end. For small tickets you understand well.
→
09
Fixing a Bug
Diagnosis-first, staying in the loop: check the cause before it codes, read the fix, then ship. For bugs where a wrong fix would hurt.
→
↳
Fixing a Bug — On the Loop
The hands-off variant of 09: hand it one goal and read the PR at the end. For routine bugs, once you trust the flow.
→
10
Spike
A timeboxed question — explore, write up a recommendation, and stop. No PR.
→
↳
Spike — On the Loop
The hands-off variant of 10: one goal researches, verifies, and writes the decision doc. Your gate is reading the citations.
→
11
Feature
Too big for one pass: break it into pieces, run the Story loop on each, then integrate.
→
↳
Feature — On the Loop
The hands-off variant of 11: the contracts stay yours; each child runs as one goal in parallel, and you read the PRs as they land.
→
12
Two Tickets at Once
A second ticket lands mid-run, same repo: give each its own working copy and its own session with a worktree.
→

Extras

beyond the core path
✦
Loop Engineering
Turn, goal, time, proactive — every way to keep the agent working, sorted by what stops the run.
→
✦
Outsource the Implementation, Not Your Brain
What stays yours as the agent writes more of the code — the decisions, the responsibility, the understanding — and how to use the agent to keep them sharp.
→
How to use it

New to this?

Request the day-one access list in Setup first — parts need your team lead. Then start with How Coding Agents Work (00) and take the chapters in order, 00 → 12; the full install can wait until Chapter 03 first uses the tools.

Got a ticket in hand?

Jump to the playbook that matches it and pull the one or two foundations you need. Running the Playbooks (07) tells you which one fits, and Words & Commands defines every name the pages use, one line each.