Sit two of your competitors down with the same AI model, hand over identical market data, and ask each of them to write a growth strategy. You'll get back two documents that could have come from the same hand: same priorities, same target segments, same tidy three-year plan with the same confident bullet points. That's roughly what's starting to happen across whole industries now that everyone runs their strategic thinking through the same handful of models.

Same inputs, same answers

This cuts against most of what you'll read about getting ahead with AI: the tool itself has stopped being the advantage. When we all point the same model at the same data, we shouldn't be surprised that we all reach the same answer. Harvard Business Review gave the pattern a name in May — the “agentic convergence trap” — when three strategy academics argued that competing firms plugging identical tools into the same public data watch their decisions drift towards a single point, and the differentiation that built each business slowly disappears. They're careful to add that this is a leadership failure, not a technology one. The machine does exactly what you ask of it; the trouble is that you and your rivals are all asking it the same thing.

So “we use AI and they don't” is an edge with a shelf life measured in months. What's left, once everyone has the same brilliant assistant, is everything the assistant can't reach: your judgement, your taste, and the odd, hard-won data that only you happen to own.

Picture three plumbing firms in the same town, each asking the same model how to grow. They'll all be told to specialise in bathroom renovations, raise their prices, and chase Google reviews, because that's the sensible advice the data supports. None of it's wrong. The problem is that it's now everybody's advice, which means it has stopped being a strategy and become a baseline. The moment a move is obvious to the machine, it's obvious to every competitor running the same machine.

The moat moves to what the machine can't reach

Some people have already worked this out. Roger Lynch, who runs Condé Nast, has been telling his teams to plan as if search traffic will fall to zero, and to treat the flood of AI-generated content online as a gift rather than a threat. His reasoning is that when the web fills up with competent, machine-made sameness, genuinely human work — original reporting, named experts, a point of view with a real person behind it — becomes more valuable. There's nothing sentimental about Vogue and The New Yorker putting more into human editorial. It's the one thing the machines can't flood the market with.

So if the tool is no longer the moat, what is? Three rather old-fashioned things, which is precisely the point. Judgement: knowing which of the model's plausible answers is actually right for your business, and being willing to overrule it when your experience says otherwise. Taste: the harder-to-pin-down sense of what's good, what fits, what your particular customers will genuinely love. And proprietary data: the strange, specific, first-party stuff that lives in your business and nobody else's. Your service logs. The complaints your customers actually make. The patterns you've spotted over fifteen years that have never been written down anywhere a model could scrape them. The firm that wins is usually the one whose owner reads the model's tidy recommendation, decides “that's not us”, and backs a weirder bet the data would never have surfaced: a segment everyone ignores, an offer that looks inefficient on paper, a way of doing things no competitor would think to copy.

For most of us, this means the model matters far less than the context we wrap around it — the questions we think to ask and the judgement we apply to whatever comes back. Ben Thompson makes a version of this point on Stratechery: as the model becomes the commodity, the company wrapping judgement and data around it ends up with the advantage. Which is really the old truth in new clothes. I've argued before that AI doesn't replace thinking, it replaces knowing, and the convergence trap is that same point scaled up to a whole market. Knowing things stopped being scarce years ago. Knowing what to do with what you know never did.

This is also why “made by a human” is becoming a genuine selling point. I've written about how visible signs of a real author — even the odd typo — are turning into a trust signal in a world of frictionless machine polish. Strategy is heading the same way. When every competitor's plan reads as though it came off the same production line, the business with an actual, idiosyncratic point of view stands out simply by having one. How long that gap lasts is a fair question — models keep getting better at mimicking a voice, and I wouldn't bet on any moat being permanent — but it's real today, and wider than most owners assume.

Competing on what doesn't come out of the box

None of this means ignoring AI. Use it; it would be daft not to. But use it knowing it gives you parity, not advantage, and spend your real effort on the things that don't come out of the box. A few practical places to start:

  1. Feed it what nobody else can. The public data is shared; your first-party data isn't. Get your service records, customer conversations and operational history into a shape the model can use, and you've handed it raw material your competitors simply don't have.
  2. Argue with the output. Treat the first plausible answer with suspicion, because the obvious answer is now everyone's answer. The interesting strategy usually lives in the place where you disagree with the machine and can say exactly why.
  3. Protect the human layer. Taste, brand, the specific feel of how you do things — these aren't inefficiencies to be automated away. They're increasingly the only part of the business a competitor can't buy off the shelf.
  4. Measure how far you diverge. HBR suggests treating divergence from the obvious as a strategic signal in its own right. If your plan looks like everyone else's, treat that as a warning.

The businesses that come out ahead over the next few years won't be the ones with the best AI, because everyone will have that. They'll be the ones who remember that a tool everybody shares can't, by definition, set you apart, and who put their energy back into the judgement, the taste and the hard-won knowledge that no model can hand a competitor for free.