Results · My own working system
An AI operating system for 22 projects
How I stopped re-explaining myself to an AI, and what I'd set up for your team.
The problem
So, honestly, I had about 22 projects on the go: client work, products I'm building, a recipe site, a planner, training material. I work in short windows, and every time I came back to one, I spent the first chunk re-explaining where I was. To an AI, that meant starting from scratch every time. To me, it meant losing the thread.
The idea
I stopped treating the AI like a chatbot and started treating it like a new team member. A new hire gets a handbook, a few playbooks for the repeat jobs, and a routine. So I built those for the AI, in four layers.
What I built
1. The handbook
One global instructions file covers how I work, tone, privacy, and when to ask before acting. Every project also has an index and a docs folder, so a fresh session starts in the right place without me explaining anything. 21 of my 22 projects have one.
2. The playbooks
13 skills that turn repeat work into one command: requirements, then spec, then plan; a wrap-up check; a quiz to test my own understanding; drafting in my own voice; a daily planning pass.
3. The automation
Two checks run on every message. One confirms the working folder so I never edit the wrong project; the other offers my daily planning once a day.
4. The data
A small script ranks every project by how long it's been quiet. An inbox sends parked tasks into my planner app, so "for later" doesn't mean "forgotten". Memory notes carry decisions from one session to the next.
Today's Three ties it together
Once a day it offers at most three suggestions: one update, one new thing, and one thing that has dropped off. They're sized to my energy that week. It's an offer, never a to-do list, and it learns from a log of what I actually took.
The guardrails
This is the part I'd insist on for any client team. Private projects show only dates, never contents. There are hard rules on what never leaves the machine. Nothing goes public, costs money or is sent outward without my yes. The AI can move fast because the boundaries are written down, not remembered.
What happened
21 of 22
projects start from a written handbook
150
sessions across 18 projects on 27 days
3 min
median time to a first change, from my first message
62% of sessions made their first change within 5 minutes. Parked work comes back through Today's Three and the inbox instead of living in my head.
A note on those numbers: "first change" is my stand-in for time to resume, measured from session logs by a script I wrote to count it. It isn't a stopwatch, and the logs only go back about five weeks, so treat them as minimums.
Who did what
I'll be upfront: this was built with AI. Claude wrote the scripts and the first drafts. I designed the system, wrote the rules, chose the priorities, and reviewed what it produced. That split is the whole point; the AI does the typing, and I own the decisions.
Skills this shows
Process and systems design, automation, data-informed prioritisation, and working with AI well: context, docs, prompts and reviewing its work.
What I'd do next
A version your team could adopt. Same four layers, set up around your tools and your rules: a handbook for your projects, a few playbooks for your repeat work, and guardrails you've signed off on. If your team keeps re-explaining things to an AI (or to each other), let me know and we can talk it through!
Want a system like this? Let's talk.
Bring one problem to a free 20-minute call.