Things I’ve made
How I run my work
I run a small business and a content channel. Everything behind them sits in one system of notes, rules and checks, so nothing important lives only in my head.
Business
Content
Personal
An AI assistant works inside every area.
- RulesHow work in that area is done: the voice to write in, the steps to follow, what never to do.
- MemoryWhat has been decided, and why, so a choice made last month isn’t argued again.
- Current stateThe one place each fact that changes is kept, such as a link or a price. Every other note points there, so no two notes can disagree.
How I use AI to do it
An AI assistant works inside the system every day. Before it starts, it reads the rules and memory for that area, so it picks up where we left off instead of starting from nothing. I decide what gets made; it drafts, builds and checks.
- Drafting and building. Web pages, documents, scripts and the small tools behind them, which I review before they’re used.
- Reading the numbers. Ad results and website recordings go into tables and through the maths, so a decision rests on the data rather than a feeling.
- Checking. Important work gets a second read against its sources before I rely on it.
- Learning from mistakes. When something goes wrong, the fix and the lesson are written into a note, so the same mistake isn’t made twice.
How I keep it in check
An assistant with access to my files needs limits that hold even when I’m not watching. Every change starts with a plan, keys stay encrypted outside the system’s files, and every change ends with the same seven checks. Here’s how that has held up so far.
- Graded186changes to the system, each one checked before it counted as done
- Checked60+nights the whole system has been checked, since August 2026
- Tested245safety tests guarding the jobs that run on their own, all passing
- Learned140lessons written down, so the same mistake is caught next time
Counted in October 2026. Changes graded since September 2026.
The seven checks are yours to take. Here’s what’s in the download:
1 · The checklist
The System Change Scorecard
Seven checks to run before calling any change to your system done: one home for each fact, checked against the real source, watched, updated everywhere, easy to undo, written down, and safe.
2 · The gate
Five checks that can’t be skipped
If any of them fails, the change isn’t done, however good the rest looks. It stops “nearly finished” from passing as finished.
3 · The log
Three verdicts and a one-line log
Ship, ship with a follow-up, or not done. Keep one line per change, and within a few months the log shows where your system keeps slipping.
4 · The example
A change graded line by line
A weekly job that drafts invoice reminders, scored on all seven checks, so you can see what a pass and a follow-up look like before you grade your own.
5 · The AI safety check
Locks, not instructions
An AI that reads emails or web pages can be steered by instructions hidden in them. Check 7 makes sure anything that runs on its own is unable to do harm, not just told not to.
Download it from my GitHub
Free under the MIT licence: copy it, change it and use it in your own system. More parts of my system will follow there.
If it helps you, give it a star on GitHub. It tells me it was useful, and helps other people find it.