AI First Principles is the way I’ve worked since 2020, and the whole of it fits in four words: AI first, human second.
It means the AI carries the work and defers to a person only when it genuinely must. That’s the opposite of how most people use AI. They keep a human in the loop at every step and treat the AI as a glorified autocomplete — a clever assistant that has to ask permission before it does anything real. I flipped that. The AI leads: it plans, decides, builds, tests and ships, running on its own for as long as it usefully can, and a human only steps in where human judgement actually changes the outcome.
That single inversion is the difference between AI as a toy and AI as a workforce.
Where it came from
Back in 2020, while most of the industry was still debating whether AI could write a usable function, I was rebuilding how I work. Twenty-five years in IT, infrastructure, web and systems had taught me exactly what to build and where things quietly break. The missing piece was leverage — a way to do far more without simply hiring a bigger team. This way of working became that leverage. Five years on, it isn’t a theory; it’s a proven operating model that runs my businesses.
How it works in practice
The principle rests on three rules:
- AI leads, humans don’t gate. Give the AI the context, the standards and the authority to make real decisions — then let it act, rather than approving every step.
- Let it run longer before it asks. Most value is lost the moment you interrupt. The further the AI can run on its own before it needs a person, the more it produces. So the discipline is to design the work so it can run long — and only escalate to a human when it genuinely must.
- Human second, not human-never. People still matter — for strategy, taste, judgement and the calls a machine shouldn’t make alone. The point isn’t to remove the human; it’s to put them in the right place: second, not in the middle of everything.
What actually makes it work: structure, not memory
Letting AI lead only works if it doesn’t drift — and AI drifts by default. The longer you let it run, the more that matters. So the real craft isn’t the prompt; it’s the structure you build around it. Anything that has to be guaranteed can’t live in an instruction the AI is asked to remember — it has to live in the shape of the system itself. Get that right and the AI can run far longer, on its own, without going off the rails.
- One source of truth. The facts — who the work is for, the standards, the settings — are set up once in a central registry the AI reads and never re-types. Set it up once; every run points at it. Nothing drifts, because nothing is copied.
- Skills that say where things are and how to use them. Reusable, properly-documented capabilities the AI picks up only when it needs them — so it always knows where to look and how to act, instead of guessing.
- A flat structure it reads from the top down. No deep hierarchy. The AI reads up the chain of folders, so the folder it’s working in tells it exactly what it’s doing and which rules apply. It can’t lose its place, even on a long run.
- The cheapest tool that does the job correctly. Logic goes in code, facts go in configuration, and the AI is left to do only the genuine judgement. Most of the work never touches the model — which means most of the work can’t drift.
That foundation is what makes AI first, human second safe to actually do, rather than a nice idea that falls apart in production.
Why most people can’t run it
It only works if you can tell when the AI should lead and when a human must take over — and that judgement comes from operational grounding. Most AI people lack it; most experienced IT hands aren’t AI-native. I’m both — twenty-five years in IT and systems, five years deep in AI. That combination is the unlock, and this principle is how I put it to work.
What it produces
This isn’t advice — it’s output. Running on AI First Principles, I do a minimum of ten years of work in thirty days, every month — and I mean that precisely: ten years of a single person’s man-days, compressed into a month. It isn’t one big multiplier; it’s three things compounding. Admin that takes a human the best part of an hour — wiring a contact form across a mail service, DNS and environment variables, then verifying it — is actioned and live in about five minutes. Development multiplies on top, because several AIs run on the same project at once. And it holds, because the standards are applied consistently on every run, where people simply aren’t.
That’s how software the market leaders charge thousands per user per year for gets rebuilt better and taken further in a fraction of the time; how legacy systems get remade modern and connected; and how the tooling and infrastructure I needed and couldn’t buy got built — because I built it myself.
AI first, human second. Let the AI carry the work and run as far as it can — and bring the human in only when it truly matters.
A note on the two spellings
You’ll see it written both ways — AI First Principles and AI-First Principal. They’re the same methodology, and this page is set up so search engines treat both as one.