claude
a3c2872526
Four real-capture terminal screenshots covering the last of the automatable
blog placeholders:
- m12-reflog.png: real git reflog on a demo repo where an agent's
reset --hard lost a "6h of work" commit. The lost
commit line (HEAD@{2}, feat: add count command) is
highlighted amber so the reader's eye lands on the
recovery target.
- u4-mcp.png: the tasks MCP server actually queried via stdio
JSON-RPC (real tools/list response). Shows the
project's .mcp.json, then claude mcp list marking
tasks as Connected, then a tool list with the real
tool signatures (list_tasks() and add_task(title:
string)) with their live descriptions from the server.
- u4-orient-diff.png: two side-by-side terminals. Left: real head -18 of
ORIENT.md produced by the M23 lab's orient.py against
a small quotient demo project. Right: a real, tightly
scoped git diff on src/quotient.py adding a
divide-by-zero guard. Facts first, then one small
change, exactly M23's teaching.
- u5-reviewer.png: real python3 reviewer.py apply ai-review.sample.json
from the M24 lab. Shows the reviewer's output: summary,
three findings sorted by severity (BLOCKER in red,
SUGGESTION in amber, NIT dim), the agent's
recommendation (request_changes), and the explicit
"human decides" gate section.
Uploaded to WP as media 17197 (M12), 17198+17199 (Unit 4), 17200 (Unit 5).
Embedded in drafts #17132 (M12), #17136 (Unit 4), #17138 (Unit 5), all
scheduled: M12 Thu 2026-08-06, Unit 4 Thu 2026-08-13, Unit 5 Tue 2026-08-18.
Unit 3 (#17134, Tue 2026-08-11) and Capstone (#17140, Thu 2026-08-20) are
scheduled and awaiting Justin's screenshots of a real red-CI PR (recklessop/
m11-worktrees-demo#5) and a real green-CI PR (recklessop/m11-worktrees-demo#4)
respectively, both set up by this batch.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TfzV5QvtPDz8LJS3Pu5VLT
The Workflow
The Toolchain Around AI Coding
A living course for IT professionals who are comfortable in an AI chat window and starting to build real software with it, but who are still copy-pasting between the chat and their files. The goal is to replace that loop with durable engineering workflows: version control, collaboration, CI/CD, runners, and the tools that extend AI into real systems.
Thesis: the model is the cheap, swappable part. The workflow around it is the skill that lasts. This course is deliberately model- and vendor-agnostic: whichever LLM you use, the scaffolding is the same.
This repo is the course, and it also dogfoods the course: it's version-controlled, it commits its
own AI instructions file (AGENTS.md, the subject of Module 5), and each module is
built on a branch and merged through review, the same motion the modules teach.
Read it as a book
The lessons render into the Wiki as a
navigable textbook (unit-by-unit sidebar, one page per module, prev/next links). The wiki is
generated from modules/ and kept in sync automatically; it's build output, so read it there but
edit the lessons here in modules/. See tools/ for the generator and the sync
workflows.
Who this is for
IT professionals who are fluent in an AI chat window and comfortable with ops concepts. Not beginners. If you already paste code between a chat tab and your editor and feel the friction, you are the audience. You will not be taught what a variable is; you will be taught the engineering scaffolding that makes AI-assisted work safe, shareable, and repeatable.
How the course is built
It's a dependency chain, not a topic list. Every module assumes only what the previous ones taught, and nothing references a tool before it's been introduced. The 27 modules group into five units, plus a capstone finale.
| Unit | Modules | Theme |
|---|---|---|
| 1: Get out of the chat window | 1–7 | The local foundation: version control, committing the AI's config, getting the AI editing real files safely. |
| 2: Make it shareable, reviewable, recoverable | 8–12 | The team layer: hosting, issues, review, collaboration, recovery. |
| 3: Automate the checking and shipping | 13–19 | The pipeline: tests, CI, security scanning, containers, secrets, delivery, runners. |
| 4: Extend the AI into your systems | 20–23 | The frontier: MCP, skills, securing them, existing codebases. |
| 5: AI in the loop | 24–27 | Agents inside the pipeline, from assistive to autonomous, plus the evals that make it trustworthy. |
| Capstone | finale | One real feature taken end to end. |
Durable core vs. expansion zone. Modules 1–14 are the stable foundation. From Module 15 onward is the expansion zone, where a fast-moving space keeps handing us new lessons. Volatile material lives toward the back so the front stays stable as the course grows.
See the-workflow-syllabus.md for the full module-by-module plan and
the reasoning behind the sequencing.
How git works in this course
You don't memorize git commands here. Modules 1–3 have you run the basics by hand so you build intuition (the AI is still in a browser chat). Module 4 puts the AI in your editor/CLI, and from there you direct the AI to do the git work (commit, branch, merge, revert) and verify the result. Think arithmetic by hand first, then a calculator. You learn that git is critical and how it works; the AI drives the keystrokes.
Format and conventions
- Written lessons + interactive labs. Every module is a README you read and a lab you run at the keyboard. There are no quizzes; there's a "you're done when…" check.
- Run labs on your own machine, any OS. No sandbox or cloud account required. Where a lab needs code, it leans on Python or shell, picked per lab, kept as small as possible. The concepts are language-agnostic; the labs just need something concrete to run.
- Claude Code as the worked example. Commands and labs use Claude Code as the concrete agent
(
claude --version # sub your own agent); the concepts stay model- and tool-agnostic. - GitHub is the default, not the requirement. Hosting examples use GitHub because nearly everyone will encounter it, but the course is provider-neutral and includes an optional self-hosted-forge track for on-prem and air-gapped environments.
- Self-checks only. No grading, no certification; each module ends at a concrete done-criterion.
Repo layout
ai-workflow-course/
README.md # this file
AGENTS.md # committed AI instructions; dogfoods Module 5 (vendor-neutral name)
the-workflow-syllabus.md # the full course plan (source of truth for structure)
_TEMPLATE.md # the shape every module follows
modules/
01-the-copy-paste-problem/
README.md
lab/
02-version-control-as-a-safety-net/
README.md
lab/
...
capstone/
README.md
assets/ # diagrams, images
Status
All 27 modules and the capstone are written and reviewed. The lessons render to the
Wiki as a textbook, kept in sync from
modules/ by CI. Each lab is skip-friendly: copy that module's lab/start/ snapshot into a
fresh tasks-app, commit, and run that lab without doing the earlier ones.