In telecoms, there were always two camps. Those who looked at the handset, and those who looked at the network.
The handset is what you hold in your hand. It's beautiful, it's new, it changes every year. It's what goes in the ads, what you queue up to buy. And it's what's worth nothing eighteen months later. The network, by contrast, is invisible. Nobody marvels at a switch or an antenna. But that's where the money was, the durability, the real barrier to entry. People changed phones every two years without ever changing networks. The handset was disposable. The network was an asset.
The players who won, over time, are almost never the ones with the finest handset. They're the ones who owned the layer underneath, the one every handset ended up plugging into.
I look at the AI market today, and I see exactly the same scene.
Everyone is looking at the handset. Which assistant is best this month. ChatGPT or Claude. Copilot or Mistral. Which version, which model, which context window. We're comparing handsets. And just like handsets, the answer will have changed in six months. The model your teams love today will be outdated before your next fiscal year. It isn't an accident, it's the nature of a terminal: it is replaceable. That is even its purpose.
That's where the confusion begins. We think the strategic question is “which assistant to choose,” when it's a handset question. It will come back, identical, a year from now.
A terminal makes one person faster. A network makes an entire organization capable.
It isn't the same scale, and above all it isn't the same time horizon. The value of a terminal evaporates the moment you swap it. A network's value accumulates with every use. That's the whole difference between an expense and a capital asset. The first goes out every month and leaves nothing. The second builds, and ends up worth more than all the rest.
Yet today, in most organizations, AI is purely a terminal. Everyone has their own assistant, off in their corner. What works stays in a private history, in a personal account, in a routine no one else ever sees. When the person leaves, or when the model changes, everything starts over from zero. You've paid for handsets, for months, across your entire organization. You haven't built a network.
The missing layer is precisely the one telecoms understood forty years ago. A layer beneath the terminals, that keeps what matters, makes it reusableReuseThe same practice serving many times, across many spaces, the key measure of the Practice Graph's value., and depends on no handset in particular. For an organization working with AI, that layer is its practicesPracticeA unit of know-how captured in Marylink: not a document but an executable structure (content, prompt, rules, style).. Not dead documents in a shared folder. Typed, governed, executable practices: the right sales method, the framework your experts validated, the rebuttal that works, the in-house reflex. Plugged into any assistant, through the open standards now spreading. You switch handsets whenever you like. The network stays, and it grows richer with every use instead of leaving with those who leave.
I say it all the more freely because the market, me first in my early days, long believed the race was won on the terminal. It was never won there. Not in telecoms, not anywhere. The terminal captures attention. The network captures value. And confusing the two means spending a fortune on the disposable part while neglecting the only one that compounds.
So the question I put to leaders isn't: which assistant your teams use.
It's: what's left, when they close the tab?