You do not beat an algorithm with another algorithm

This summer I upgraded to the newest AI model. It is roughly twice as fast as the previous one, and measurably cleverer. On its first working day it helped me take a decision in under an hour that would normally have needed a day. The following week, the decision turned out to be wrong. It was merely wrong well ahead of schedule.

Which is progress, of a kind. I can now be wrong twice as fast, and with footnotes.

This is the first essay in Sovereign Universe, and it concerns the one question everything here will keep circling back to. Everyone upgrades their machines. Nobody upgrades their questions. What happens to a world where the answers get faster and faster while the questions stand still? And who, in that world, defines your reality, you or whoever owns the machine?

Look around any meeting room. Everyone acquired new tools this past year. Text, analysis, images, code, all of it arrives fully phrased in seconds, and all of it looks convincing. That is precisely the problem. AI and people in power now produce convincing stories faster than anyone can check them. Some of the stories are true. Many are not. And no label comes attached to tell you which is which.

The standard remedy is more technology. Buy a better model, fit a sharper filter, hire a new algorithm to guard you from the previous one. This sounds sensible right up to the moment you see it for what it is. An arms race in which you have generously agreed to fund both sides.

So here is the house rule of this universe, and it may as well be carved over everything we make. You do not beat an algorithm with another algorithm. You beat it with method.

By method I do not mean anything of mine. I mean the toolbox civilisation has already paid for. The scientific methods we have spent two and a half thousand years developing for one purpose, testing claims. Legal method to test an argument. Marketing science to test a claim of growth. Journalistic method to test a source. Mathematics to test the sums that look a little too good. Nobody commands all of them, and nobody needs to. The work is finding the method the problem requires, and applying it properly. And the methods do not care whether the claim comes from a machine or a minister.

This is where the value is moving, in plain sight. The detail is becoming cheap. Drafts, analyses, summaries, everything the machines produce in seconds, falls in price by the month. What does not fall in price is the ability to check it. Whoever can verify what the machine produces is amplified by it. Whoever cannot has let out their judgement, and calls the arrangement efficiency.

If you run an organisation, you already know the invoice. Guesswork, filed under strategy. Plans resting on premises nobody has tested, budgets approved because everyone around the table nodded, and lately AI reports that nobody in the room can check. Method is the cheapest cost-cutting there is. It does not remove people. It removes guessing.

And if you are, above all, a person with a phone, the stakes run past money. The feed chooses what you see, autocomplete suggests what you think, and somewhere in that stream lives the question of who exactly is holding the pen in your life. We take that question up in the very next essay, and it gets personal.

Hence three tracks from here. Sovereign Enterprise, for the organisation that intends to keep its identity while AI agents and narrative power rewrite the rules. Sovereign Self, for the person who intends to keep their judgement, and themselves, in the same noise. And Sovereign Lab, the workshop both draw from, where the methods are tested and explained so that they can be used within a day. One idea per essay, never more. We name the pattern. You place your own reality in it.

The word sovereign is doing deliberate work here, and it does not mean what the internet thinks. No islands, no camp, no awakening. Sovereignty, in this universe, is a result. Verify what you are told, and you can decide for yourself. Decide for yourself, and you can negotiate. A consumer accepts the terms. A party negotiates them. The rest of this series is, at bottom, about changing chairs.

If you take one thing from this first essay, let it be the upgrade that costs nothing. The next time anyone hands you a perfect answer, human or machine, ask one question before you nod. How do we know this? Five words. You will be surprised how often the room goes quiet, and how much it pays to be the one who asked.

My model is still fast, and I am keeping the subscription, gladly. But the most important upgrade this summer was free. I swapped what does the machine say for how do we know this, and got back something I had not noticed renting out. My footing.

Just around the corner, essay two. A company has a brand. You have a self. And then we sail. Three episodes along the most famous journey home ever told, just as it returns to the cinema. Do come in. Bring questions.