The Model Is the Cheap Part. The Rail Is the Moat.

The AI industry is converging on a truth that’s uncomfortable for a lot of vendors: the model is commoditizing.

Capable intelligence is already cents and an API call away. Benchmark differences between frontier models narrow every few months. You can build a genuinely useful AI assistant in a weekend with off-the-shelf components. The thing everyone called the hard part, training a model, getting to useful capability, isn’t the hard part anymore.

So value migrates. In any market where a scarce thing becomes abundant, value moves to whatever’s still scarce around it. For AI, two things qualify: proprietary context that makes an answer yours rather than generic, and distribution, the ability to get a governed answer or action to the exact point where a decision gets made.

For desk workers, distribution was solved decades ago. The AI lands in the tools they already use, the productivity suite, the communication platform, the browser. It’s almost invisible, which is why nobody in the knowledge-worker world thinks of distribution as the hard part. For them, it isn’t.

For the frontline, the 80% of the workforce without a desk, an email, or a seat at any system of record, distribution is the entire unsolved problem. No inbox for the AI to land in. No directory to authenticate against. No portal the worker was ever given. The intelligence has nowhere to go.

That gap is the moat. Not because anyone planned it. Because closing it requires something that can’t be stood up in a procurement cycle: an accurate, continuously updated model of a workforce in constant motion. Who each person is, by badge number, clock-in ID, or role, not an email they don’t have. What they’re permitted to see, derived from attributes that change as they move. How authority flows through a delegation hierarchy that no org chart captures. How all of that stays true while people are hired, transferred, re-badged, and terminated every single day.

Building that took 14 years. It can’t be copied quickly because it’s not a feature, it’s an architecture you arrive at through direct experience, one customer and one edge case at a time.

The model is a feature. Any well-funded team can build or rent one. The rail, the distribution layer that gets a governed AI answer to the right frontline worker, with the right permissions, in their language, at the point of work, is infrastructure. It’s the moat. Organizations building frontline AI strategies will eventually hit the wall that makes that moat legible: you can build the intelligence, but you can’t deliver it without the rail.

We built the rail first, without knowing this moment was coming.