Essay
The fear in most boardrooms gets phrased as a technology problem: can a machine do this job. The sharper question is a market problem: what happens to a market when everyone in it gains access to the same capability at the same time. Quality doesn’t collapse. It converges — and convergence is a far more dangerous condition for most businesses than obsolescence.
Consider what actually happened the moment large models became widely available. Millions of people, most of whom had never written a strategy document or a piece of brand copy in their life, gained access to a tool that could produce something structurally competent in seconds. The floor of quality rose everywhere at once. What did not rise with it was the ceiling, because the ceiling was never a production problem to begin with.
Why the same prompt produces the same brand
Ask the same model the same reasonable question and, allowing for surface variation, you get answers that share a centre of gravity: similar structure, similar tone, similar list of considerations, because the model is drawing on the same underlying patterns regardless of who is typing. That isn’t a flaw. It is exactly what a well-trained pattern-matching system is supposed to do. Which is precisely why using it without a strong point of view produces work that looks like everyone else’s work.
Walk through any industry right now and you can feel it. Pitch decks that read like the same deck with the logo changed. Marketing copy with the identical rhythm of short punchy sentence, longer explanatory sentence, rhetorical question. Strategy documents built from the same five frameworks arranged in the same order. None of it is wrong. All of it is interchangeable, and interchangeable is the one thing a market will not reward with a premium.
When everyone has the same tool, the tool stops being the advantage.
Where the differentiation actually moved
It moved to the layer above the tool: what you decide to ask it, what you already know that lets you recognise a good answer, and what you are willing to change once you have it. Three people with identical access to the same model will produce wildly different outcomes if one of them understands their customer deeply, one has a genuinely unusual point of view about the market, and one is simply trying to finish the task quickly. The model did not create that gap. It revealed it, at speed, and it will keep revealing it for as long as the tool remains this good at fluency and this indifferent to correctness.
Judgement is the first component: knowing which of the model’s plausible answers is actually right for this customer, this moment, this constraint. Point of view is the second: having something to say that did not come from averaging everything that has already been said, because a model trained on the internet’s consensus will, by construction, tend back toward that consensus unless a human actively pulls it away. Taste is the third, and it is the hardest to describe precisely because it resists being reduced to a rule; it is the accumulated sense of what is good that comes from having made and judged a great deal of work before the tool ever existed.
- Does this sound like us, or does it sound like the average of everyone we compete with?
- What did we know about our customer that the model could not have known?
- Would a competitor with the same tool have produced something meaningfully different, or the same thing with different words?
The uncomfortable part for organizations
Most companies are currently rewarding the wrong behaviour. Teams get praised for shipping faster, for producing more variations, for reducing the hours spent on a deliverable, and all of that is measurable and easy to celebrate in a quarterly review. Almost nobody is measuring whether the accelerated output is distinguishable from a competitor’s accelerated output, because that requires an editorial judgement that most reporting structures are not built to make.
This is why the honest diagnosis of a homogenising market is rarely a technology audit. It is a culture audit. A team that has a real point of view about its customers and the confidence to defend an unfashionable position will use the tools to produce more of that point of view, faster. A team that has never had to develop a point of view because the market used to reward competent execution will use the same tools to produce more competent execution, and will be baffled when competent execution stops converting.
What to do before your market fully converges
Deliberately audit your own output against the category average, not against your own past output. It is entirely possible to be improving quarter over quarter while becoming less distinguishable, because the whole category is improving at a similar rate using similar tools. Improvement relative to yourself is a comforting metric and an increasingly meaningless one.
Invest in the inputs a model cannot generate on its own: direct exposure to customers, an unfashionable opinion you are willing to defend in public, institutional memory about what has actually worked and failed for your specific business rather than businesses in general. Those are the raw materials of a genuine point of view, and a genuine point of view is the only thing that survives being run through the same tool as everyone else and still comes out looking like you.
The machine was never going to replace you by being better at your job than you are. It replaces the version of you that had stopped having an opinion about the work. Everything else, it is happy to make faster on your behalf.