Most disappointing AI output is not the model being weak. It is the model working blind, asked to reason about a product it has never been allowed to see.
I have tested that claim on my own work more times than I can count. The same model that hands back a generic, useless answer when it can only see my question hands back something sharp and specific the moment it can read my code, my tickets, and my running app. The gap between those two answers is not intelligence. It is access, plus the standards and procedures you wrap around it. This course builds all three.
Who this course is for
You are a working QA engineer, and AI tools are either already in your workflow or arriving whether you asked for them or not. This is the third course in my free set. If you are coming from Test Automation Basics, you already have a test suite against my practice store, and this course puts an agent to work inside that world. If you are starting here, that is fine too. Every lesson stands on its own, and most of the daily exercises need nothing more than a free model account.
One definition before we start, because I will use the word every day. A coding agent is an AI that does not just chat: it can read your repo, edit files, run your tests, and open pull requests. Claude Code, Cursor, and Codex are the ones you will hear named most. When I say agent, that is what I mean.
The triangle this course builds
Three pieces decide whether an agent is genuinely useful on a QA team, and the middle of this course builds each one:
- A rules file (CLAUDE.md or AGENTS.md) carries what the agent always knows: your conventions, your locator rules, your fences. That is Day 4.
- A skill carries what it does on demand: a procedure like writing a test plan, encoded once, invoked when needed. That is Day 5.
- MCP (Model Context Protocol) carries what it can see and touch: your repo, your tickets, your team chat, your running app. That is Day 2.
Hold those three apart in your head and almost everything about working with agents gets simpler. Most frustration with AI in QA is one of the three missing.
The ten-day map
- Day 1. What AI can and cannot see, with the lesson I learned building a game with AI agents.
- Day 2. Real access through MCP: six sources, one file, and the prompt that proves the wiring.
- Day 3. What that access buys you: a Slack thread becomes a filed bug, then a pull request, with my exact prompts.
- Day 4. The rules file, built with the agent instead of by hand.
- Day 5. Skills: encoding the procedure you keep re-pasting.
- Day 6. Test plans with AI: risk first, cases second, published where the team works.
- Day 7. Test data and edge cases with the expected result attached.
- Day 8. Reviewing AI output without rubber-stamping it.
- Day 9. The control layer: five questions before anything ships.
- Day 10. Capstone: the full loop on a real store, then the assessment and practical review.
What you will be holding at the end
Every lesson leaves you with something you can use the same day: a prompt to copy, a file to commit, a checklist to run a real diff through. By Day 10 you will have a working rules file with a review fence, tests you reviewed and corrected, and a release note that takes a position. Your final email unlocks a 25-question competency assessment. The certificate also requires an approved practical that shows your tool choices, checkable rules, evidence, corrections, and human review.
One thing holds in every single lesson, and I want it on the table from the start: the agent does the reading, the drafting, and the typing. You make every decision that matters, including the one about what merges and what ships. That is not a limitation of this course. It is the point of it.
Try this today
Pick the last piece of AI output that disappointed you. A vague test, a generic answer, a wrong guess about your product. Write down, as a list, what it would have needed to see to get it right: the code, the ticket, the spec, the conversation where the team decided the behaviour. Keep the list. On Day 2 you connect almost every item on it.
Talk tomorrow.