Module 11: Engineering with AI
Lesson 1.6 gave you a chat window and two rules, then told you the deep end was locked until Module 11. It said the pain had a payoff date.
This is the date.
What changes here is not the model. It is where it sits. In Module 1 you pasted things into a browser tab. From here, an AI agent works in your terminal, in your repositories, on your lab: reading files you wrote, running commands you approve, and drafting documentation in your voice.
That is genuinely a different tool, and it is the one practising engineers use daily. It is also the one that can do real damage, which is why you waited.
What's in it:
- 11.1 what changes when the AI is in your terminal
- 11.2 get an agent running against your own lab
- 11.3 context: teaching it your environment once
- 11.4 the review discipline that makes this safe
- 11.5 skills: packaging the procedures you repeat
- 11.6 documentation that sounds like you
- 11.7 the security boundary, which is a real one
- 11.8 a real task, start to finish
- 11.9 journal entry
- 11.10 checkpoint
Why the wait was the point
Lesson 1.6 put it plainly: you can only safely delegate what you can verify.
Ten modules ago that sentence was a promise. Now it is a description of you.
When an agent proposes a firewall rule, you have written firewall rules. When
it drafts a playbook, you wrote playbooks by hand in Module 10 and know what
--check would tell you. When it explains a Kerberos failure, you have
debugged one.
An agent's output looks equally polished whether it is right or subtly wrong. That is the whole problem, and there is no technical fix for it. The only defence is a reader who has done the work before, and that reader is now you.
Someone who started at Module 11 would get the same suggestions, accept all of them, and learn nothing except that a tool exists which produces confident text.
What this module is not
It is not a prompt-engineering course. Prompting advice ages badly and most of it is folklore. What transfers is the working relationship: what to delegate, what to check, and where the boundary sits.
It is not an argument that you should use this. Some engineers do not, and some workplaces forbid it. The lessons below are written so that if you never run an agent again, you still leave understanding what one does, what it can reach, and what questions to ask when your employer adopts one. That is worth having either way.
Tier 1 and up. Everything here runs on your own machine against your own files. No extra virtual machines.
This site is written with these tools, by one person, using exactly the workflow lesson 11.5 describes. The war stories in this module are from building the thing you are reading.
That is not a recommendation, it is a disclosure. You should know what you are being taught by whom, and judge the advice accordingly.