In an earlier essay I argued that the businesses that win optimise with AI, not for AI. They decide what the whole effort is aimed at, put AI at the core of it, and keep a person in the loop asking whether it still feels like them. AI is an optimisation engine. What matters isn’t whether you use it, but what you’ve aimed it at, and how it gets there.

This piece is about what happens next, because once AI is wired into the core of how your business actually works, something starts to compound that a competitor can’t simply go out and buy.

The tool is not the advantage

Here’s the uncomfortable truth for anyone hoping AI itself is the edge: the model is not the moat.

A competitor can license the same models, read the same prompt guides, hire from the same pool, and be running the same tools by Friday. If your only advantage is having AI at all, that advantage has a shelf life measured in weeks. Everyone is shopping off the same shelf.

This is why bolting AI onto a business gets you so little. It buys a faster tool, and a tool is horizontal — it sits on top of the work, available to everyone, including whoever you’re trying to beat. Speed anyone can purchase isn’t a moat. It’s table stakes.

The real advantage is what the engine has learned from running inside your business, aimed at one thing, long enough to get good at it.

What compounds when AI sits at the core

When AI is built into the core of a business rather than bolted onto the edge, it stops being a generic tool and becomes a specific asset. It sees your orders, your customers, your pricing decisions, your service history, the questions people actually ask, the deals that close and the ones that fall over. Day after day it learns the shape of your business, not businesses in general.

That accumulation is the part a competitor can’t copy. They can buy the same model, but not a year of it watching your operation and adapting to it. They’d have to start from zero and wait, the same way you did — meanwhile you’re another year ahead, and the gap widens instead of closing.

There’s a name for this in strategy, and it’s worth being honest about it. It’s a data and learning advantage, and it behaves like compound interest: small at first, then quietly decisive. Businesses that pointed AI at their core early aren’t slightly ahead. They’re on a different curve.

The bolt-on, by contrast, learns nothing durable, because it was never wired into anything that matters. It answers a query and forgets. Pull it out and the business is unchanged — which is exactly why it was never a moat. Nothing accrued.

One direction is the whole point

In the first essay I made the case that Ukraine’s edge wasn’t the drones. It was that the drones served one strategy. The same logic decides whether AI compounds into a moat or dissipates into noise.

When everyone in a business aims AI at their own goal, you get plenty of local optimisation and no cumulative advantage. One person tunes for speed, another for tidy output, another just to get through the day. AI helps all of them, because that’s what it does. But nothing pulls the same way, so nothing builds.

A moat needs direction. When the whole business trains AI on one clear thing, every interaction isn’t just solving today’s task — it’s adding to a single, accumulating understanding of what the business is and where it’s going. That’s the difference between a thousand small conveniences and one growing asset. The convenience is copyable. The asset isn’t.

This is also why the human in the loop matters more as the moat deepens, not less. Someone has to hold the direction steady, asking whether the goal is still the right one. An optimiser aimed at the wrong target will compound just as reliably — straight off a cliff. Taste is still the job. AI doesn’t have it.

The gap that cannot be closed by shopping

Put the two halves together and you can see why this becomes a defensible position, not just a temporary lead.

A competitor can match your tools. They can’t match your accumulated, business-specific learning without spending the same time you did, aimed in the same disciplined direction — and by the time they have, you’ve moved again. This isn’t a fixed distance they can sprint to close. It’s a lead that compounds while they run.

That’s a moat in the truest sense: not a clever feature or a faster tool, but a structural gap that widens through use. It can’t be bought, because it was never for sale. It can only be built, slowly, by a business that decided early what it was optimising for and stayed the course.

So the question from the first essay still stands, with a second edge to it now. What is your AI optimising for — and has it been aimed there long enough, and consistently enough, that a competitor couldn’t catch up even if they tried?

If the honest answer is that your AI is bolted on and scattered in every direction, you don’t have a moat. You have a subscription. And so does everyone else.

Nigel Price is the founder of Digital Discovery Group, specialising in ecommerce strategy, digital transformation, AI-powered platforms, and managed cybersecurity services for small and medium businesses.