John Laverick

Note

Models Commoditise. Harnesses Accumulate.

The model may be the most visible part of an enterprise AI system, but the durable capability is increasingly everything the organisation builds around it.

21 June 2026 · Technology & Leadership, Building Through AI

The model gets the hype. The harness gets the results.

As AI capabilities improve, I’m increasingly unconvinced that choosing a particular model is where most organisations will build durable advantage.

Models matter enormously, of course. But they are also improving quickly, becoming more interchangeable and acquiring capabilities that were distinctive only months earlier.

The harder thing to copy is everything an organisation gradually builds around them.

The context that lets a model understand the work. The permissions that determine what it can see and do. The tools that connect it to real systems. The evaluations that tell us whether it is performing well. The human approvals that sit around consequential actions. The tracing that lets us understand what happened. The operating ownership when it becomes part of a real service.

That’s the harness.

I don’t mean a giant enterprise AI platform designed upfront to solve every future problem. In fact, I think that would miss much of the point.

The useful harness emerges through doing real work.

Build something that solves a problem. Discover that reliable access to a particular source of context matters. Make that reusable. Build the next thing and discover you need a better permission pattern. Make that reusable too. Add evaluations because a failure taught you what needs measuring. Add a human gate because experience showed you where judgement matters.

Over time, the organisation accumulates capability.

The first workflow may be relatively expensive because much of the surrounding machinery has to be created with it. The tenth should benefit from everything the previous nine taught you.

That is why I like the distinction between models and harnesses.

Models commoditise. Harnesses accumulate.

And the harness isn’t simply there to stop the model doing the wrong thing.

It is how we encode what the organisation has learned about doing the work well: the context that matters, the standards that apply, the actions that are permitted, the things that need checking and the moments where a person needs to decide.

A harness is organisational knowledge made executable.

That feels like a much more durable enterprise asset than whichever model happens to be at the top of the leaderboard this month.

Originally published on Threads.