Today Satya Nadella, CEO of Microsoft, published a piece on his blog titled "A frontier without an ecosystem is not stable," where he lays out how he pictures his company in an economy driven by AI. What he says, in short, is that it isn't about which model is smarter, but about who owns what.
His thesis is simple: the value isn't in picking the best model to build your solutions on top of, but in owning the learning loop you assemble on top of the models. He even proposes a test for whether you really control your part: being able to swap one general-purpose model for another without losing the "company veteran," that tacit knowledge your organization has accumulated inside its own system.
I agree. So much so that I've been doing it for months. Only not at the scale of a multinational, but at the scale of one person and a VPS at Hetzner. Seen from down here, it looks like this.
My assistant, Hermes, isn't married to any model. Today, as I write this, it runs on Claude, but tomorrow I can switch providers with a single line of configuration. What doesn't change is the layer on top: my skills, the procedures I've distilled; my memory, what it knows about me and how I work; my pipelines. All of that lives in open formats, on my infrastructure, outside the model. The "company veteran" Satya talks about is, in my case, myself, encapsulated in that layer.
It isn't theory. Just today, in fact, Hermes's web tools were down. The model had no idea how to get around it, but I had the plan B documented in a skill: fall back on curl, on the browser, on a couple of alternative endpoints. The work went on without missing a beat. The knowledge of how to manage wasn't in the model but on my side of the equation.
There's a line of his I'm keeping whole: you can delegate a task, even a whole job, but you can never delegate your learning. It holds for a company of fifty thousand people and it holds for one person with a server. If you outsource the task and on top of that let the learning stay in someone else's model, you haven't saved work. You've given away your advantage.
Satya also writes that human capital "only becomes more valuable" as AI grows. Let's hope so. But what I don't doubt is that owning your learning loop doesn't guarantee that we all win. It's, at most, the condition for not being the one who loses. Sovereignty doesn't hand out the future fairly on its own; it only gives you a place to keep standing when the ground moves.
That's why I insist, post after post, on the same things: open formats, your own infrastructure, modularity. Not out of textbook purism, but because it's the only version of this story in which the company veteran is still yours when the model of the day changes. And a lot of models, and a lot of other things, are going to change.