Day 15 of 14 — the bonus one. Pete Gypps, COR Intelligence.
I've spent two weeks laying out a way of building AI systems — put the guarantees in the structure, connect the data, make it visible, keep the human for the strategy. And the honest thing to say, before anyone hands me more credit than I'm owed, is this: I didn't invent almost any of it. I connected it. The hard parts were done by people mostly long dead, in fields with nothing to do with software, decades or centuries before anyone typed a prompt.
Naming them isn't false modesty. It's the opposite — it's the strongest thing I can say for the method. Something only I believe is an opinion. Something serious people keep landing on from unrelated directions is closer to a law. So here are the ones I lean on, and what each actually gave me.
Karl Popper — a theory has to be able to be wrong
Popper's line in the sand, in the 1930s, was simple and brutal: a claim that can't be disproved isn't science, it's belief. If there's no experiment that could ever show it false, it explains nothing, because it forbids nothing.
That's the test I hold my own method to. "A model update should never break this system" is a real claim, because it's falsifiable — swap the model, and if something breaks, I was wrong. "Nothing is lost silently" isn't a slogan in my work, it's an assertion that runs and goes red when it's broken. The moment a method can't be caught out, it stops being a method and becomes a religion. Popper is why I built mine to be catchable.
The Marquis de Condorcet — independent agreement is evidence
In 1785, Condorcet worked out something about juries that turns out to be about truth in general. Take a group of independent people, each a bit more likely than chance to be right, and the majority gets more and more certain to be right as the group grows — provided their mistakes are their own.
That last clause is everything. It's why I take it seriously that a filesystem-first methodology, and Google's data-documentation framework, and the people building the best agentic tools, and folder-based note-takers all arrived at the same conclusion without coordinating. Four independent juries, four different rooms, one verdict. Condorcet is why that convergence reads to me as evidence and not coincidence — and why I'm careful to say I'm one of the four, not the source of it.
Albert Einstein — build from axioms, and prize what stays the same
Two things from Einstein, and neither is the famous one.
First: the shape of a good theory. You don't start with a thousand observations and hope a pattern falls out. You start with a small number of postulates you're willing to stake everything on, and you derive the rest. My whole method comes down to two: guarantees live in shape, and nothing is lost silently. Everything else falls out of those. That's not me being grand — it's me copying how the strongest theories are built.
Second, and more useful: invariance. The heart of relativity is that the real laws are the ones that hold no matter which reference frame you measure from. The frame changes; the law doesn't. That is exactly the test for whether you built your system at the right layer. If a new model — a new frame — changes how well the work gets done but never whether your guarantees hold, you found the invariant. If a new model breaks everything, your "laws" were just artefacts of one frame. Einstein gave me the sharpest question I ask of any system: what here holds no matter which model I use, and what's just an artefact of the one I happened to pick?
Douglas Engelbart — augment the human, don't replace them
In 1962, before almost any of this existed, Engelbart wrote Augmenting Human Intellect. His argument was that the point of the computer wasn't to automate humans out of the loop — it was to amplify what a human mind could do. The productivity gains people expected from automation kept disappointing, he said, precisely because they tried to replace the human process instead of augmenting it.
Sixty years on, that's the whole design of what I build. The machine holds the structure and moves through it tirelessly; the human does the one thing no model supplies — deciding what matters. AI first, human second, but the human is the point, not the passenger. Engelbart called it half a century before I needed the phrase.
Andrej Karpathy — the line that started me
And then the modern one, the throwaway comment that set the whole thing off: LLMs don't get bored, don't forget to update a cross-reference. I've written before about how that named my problem from the wrong side — if the machine doesn't forget, why do our systems still rot? — so I won't repeat it. But it belongs in this list, because it's the spark that made me go looking, and everyone above is who I found when I looked.
Why I'm telling you this
There's a version of thought-leadership that pretends every idea is new and self-generated. I've never bought it, and I buy it less the more I read. The people worth listening to are almost always the ones who can tell you exactly whose shoulders they're standing on, because knowing your lineage is how you know your idea holds weight and isn't just clever.
So: Popper for the falsifiability, Condorcet for the convergence, Einstein for the axioms and the invariance, Engelbart for keeping the human central, Karpathy for the spark. They poured the foundations. I did the connecting and the extending — pushing a static idea into a living system with a time axis, turning the laws into tests you can actually run, building the instrument that applies them. That part's mine, and I'm still building on it. Knowing exactly whose work I'm standing on doesn't make my own smaller. It's the reason I trust the ground under it.
Pete Gypps — COR Intelligence.
