Harness,
Don’t Drift.
A free course, taught by whatever AI you already use
Open the folder in your AI tool, say one sentence, and the model becomes your tutor. Seven lessons, six artifacts, all of it built from a project you actually care about. You leave still the author of your own work.
Nothing to install. Paste this into any chat model.
Open chatgpt.com, claude.ai, gemini.google.com, or any chat that can open a link. Paste this and send it. The model reads the course from this site, walks you through the three doors and the lessons, and hands you your work to save.
Fetch https://mindwright.ai/course/START-FROM-URL.md and do exactly what it says. (The same course is on GitHub at https://github.com/ignisspindler/you-are-the-harness.) You cannot read or write files in this session; I will save your outputs myself. Start the course now.
Every file in the course is served as plain text under mindwright.ai/course/, a live mirror of the repository, so a model doesn't need GitHub access to read it. The download route below gives you the same course with autosave and a workbook on your own machine.
The AI is the instrument. You are the author.
I built this because I kept watching the same thing happen to smart people. They sat down with an AI, got something that sounded right, and walked out of the session with their work quietly pointed somewhere they never chose. Nobody decided that. It drifted. The fix is a craft, and a craft can be taught.
So I wrote the course as a folder. No app, no login, no videos. You download it, open it in the AI tool you already use, and say: Read AGENTS.md and start the course. Your AI reads the tutor contract inside and teaches you from it, one Socratic conversation at a time, writing what you build into a workbook that stays on your machine. It takes about seven and a half hours, spread out however your life allows.
It runs anywhere a model can read a folder: Claude Code, Codex, Cursor, Gemini CLI, and whatever ships next month. If all you have is a browser chat, there is a paste-and-save routine for that too. The tool was never the point. What you do with it is.
Governance is atomizing. That leaves you.
Here is the prediction behind this course, and I'll say it plainly. Ethics and governance of AI are going to become impossible at every level above the individual. The surface is too multidimensional. The solutions an AI can generate for one person in one afternoon are too many for any manager to supervise, any regulator to foresee, or any governance expert to write a rule for. The overseers will fail, because they cannot predict all the ways AI generates new knowledge for the people using it.
If I'm right, AI interactions atomize. They stop being visible from above at all. And the only governance left standing is the kind you run yourself, on your own relationship with the AI in front of you: what you delegate and what you keep, how you check what comes back, what never gets pasted in, how the work stays pointed where you pointed it. It's already happening. This course is how you get good at the part that's yours, because nobody else is positioned to do it for you. The glossary has the full statement.
One loop. Every lesson lives somewhere on it.
Semantifacturing. The emergent human craft of AI-enabled execution: the skill of scaffolding language precisely enough to engineer one’s intent into tangible outcomes through AI tools.
Eugene J. Geis, PhD · Glossary
Your intention above the AI. Your judgment below it.
Three things stay true no matter how capable the models get. The AI is not the owner of the project. The AI is not the final judge. The AI is not the source of purpose. Those three lines are the design. Everything inside them is fair game, and the whole course is about making that inside as useful as it can possibly be while the lines hold.
Seven lessons. Six artifacts. One file that travels.
Every lesson ends with something written down, in your words, about your project. The tutor asks, you answer, it structures, and it reads the result back to you before anything is saved. Here is what you walk out with.
| Module | You leave with | Why it matters |
|---|---|---|
| 00 | The governing model, and a project to work on | Nothing here is abstract. You bring real work or you borrow some from a starter list. |
| 01 | A Project Constitution | What you are doing and what must not change, written down before the AI touches anything. |
| 02 | A Context Stack | The environment your AI works inside, and the short list of what never gets pasted in. |
| 03 | A Six-Attack Synthesis | Six ways to generate broadly without handing over the synthesis. |
| 04 | An Epistemic Airlock | How you decide what enters your project, and how you catch the answer that fits you too well. |
| 05 | A Change Ledger | How you evolve the work deliberately instead of accidentally. |
| 06 | A Harness Card | Your complete operating protocol on one page. |
And one more: your intent file, a behavioral contract you write during the course and copy into your own projects afterward, so every AI you work with reads it before it reads anything else.
Five words, in the order they arrive.
- SaferYou know what never goes into an AI tool, and you can tell a confident answer from a correct one.
- AlignedThe work stays pointed where you pointed it. Your intent file makes sure of that, session after session, tool after tool.
- ConfidentYou accept and reject on purpose, with a reason you could say out loud.
- EfficientLess re-explaining, less re-doing. Your context gets built once and travels with you.
- ProductiveYou finish things. Real things, out in the world, where reality gets to answer back.
Three steps, and the first one is a download.
Only have a chat window? Skip all three steps and use the prompt at the top of this page. If your chat can't browse the web at all, reference/no-ai-tool-nearby.md inside the folder has the paste-and-save routine.