Welcome aboard

"Thanks a lot for stopping by and trying this taster. I sincerely hope you will learn something in this course. My hope and goal is to make it approachable and easy to get started writing better code using Agentic Development. If you have feedback or questions to the content or the course itself, do not hesitate to reach out to me!
Do you know why Java programmers wear glasses? Because they don't C#"
Carsten Lützen
This course is about getting a coding agent to write code you can still live with in six months. You don't learn that by reading. You learn it by watching an agent work in a real codebase, and then changing what it works in. So Getting started has one job: get you a codebase and an agent that can run its tests. About 30 minutes, less if you have the tools already.
Your machine, your risk
You'll let an AI agent read files and run commands. It can delete things, change code you didn't ask about, or send code to the company behind the model. Work on a branch, commit before each task, and keep secrets out of reach. You follow the course at your own risk.
The full small print
You do everything in this course on your own machine and at your own risk. The course, the practice codebase and the templates are provided as they are, without any warranty. To the extent the law allows, I am not liable for lost or damaged data, lost files, leaked code or secrets, downtime, or any other loss that comes from following the course or using its material.
A few habits keep the risk small, and they are part of working well with agents anyway:
- Work on a branch, and commit before you hand the agent a task, so Git can undo anything it does.
- Keep secrets out of reach. Don't give the agent access to production
databases, real customer data, API keys or
.envfiles you can't afford to have leaked. - Check before you use a work codebase. Make sure your employer allows AI agents on it, and which ones.
- Read what the agent asks to run before you allow it, especially commands that delete, install or push.
Is this for you?
Yes, if you build software with a coding agent, or want to, and you want what it writes to stay easy to change. You don't have to be a seasoned developer. The agent writes the code. Your job is the environment it works in: the rules it follows, the checks it must pass, and the size of the tasks you give it.
You need to:
- paste a command into a terminal and read what it prints;
- edit a text file by following an example;
- say what you want from your code: how things are done, and what must never happen.
You don't need to read PHP to get started. The checks tell you whether the agent's work holds up, and Git shows you what it changed. Reading code helps more the further you go: in Part 1 you review the agent's plans and small changes, and in Part 3 you judge the tests it writes, with a review checklist that tells you what to look for. A fresh agent session can review alongside you.
You don't need to know Git or automated testing well either. The course teaches the parts it uses.
The examples use PHP and Laravel. The ideas work in any language, and a Python track may follow. Tell me if you'd want one.
How the exercises work
Every exercise has the same rhythm:
- Set something up in your codebase: a rule in
AGENTS.md, a check, a limit. You paste commands and examples from the page, or hand the setup to your agent with a prompt. - Give the agent a small, real task. On FridgeLister the tasks are ready-made. On your own code, take one from your backlog.
- Watch what it does. Did it follow the rules? Did it run the check,
and fix what the check found?
git statusshows which files it touched. - Keep what works. Commit it. When you have to correct the agent, write the correction down so you don't have to make it again.
Each exercise ends with a "Done when" list, so you know when you're finished.
How the course is built
- Getting started (this part): a codebase and an agent that runs its tests.
- The free taster (1 hour): the seven layers of a quality harness, and
your first
AGENTS.mdand check command. - Part 1: Habits. What the agent should know, what done means, small tasks.
- Part 2: Measure and guard. Numbers and guardrails.
- Part 3: Going deeper. Complexity, architecture, CRAP scores and mutation testing: the theory that lets you choose limits you can explain.
Next: pick your path.
Kept in this browser. Sign in to keep it on every device.