Taste is the New Moat

In a world where everyone has access to the same models, the same APIs, the same tools — and they do — the question of competitive advantage has to be answered somewhere new.

It used to live in data. Then it briefly lived in model size. Then it tried to live in fine-tuning, and then in retrieval, and then in agents. Each of those was a real thing for about eighteen months, and then it commoditised.

There is one place left where it cannot commoditise, and that place is taste.

Taste, in the sense I mean it, is two things. First: knowing what's worth doing. Second: knowing when something is finished to a standard you'll put your name on. Neither one is in the model. Both are in you.

Consider what happens in a typical AI workflow. You give the model a task. It produces something. You accept it, edit it, or throw it away. The model does not know whether the output is any good — it has no opinion. You know. The accept-or-reject decision is where taste lives, and it's the highest-leverage move in the entire system.

Before AI, taste was hidden inside expertise. A senior copywriter wrote good headlines. You didn't see the taste — you saw the headline. The bad headlines never made it out of the writer's head.

Now the bad headlines are generated in two seconds and presented to anyone with a keyboard. The work of choosing the good one is suddenly explicit. The person who can sift forty AI-generated headlines and pick the two that actually work is doing the same job the senior copywriter always did — they're just doing it visibly, on outputs that were free to produce.

AI doesn't replace the writer's taste. It exposes it as the actual product.

I have started using a simple test when I look at a team trying to deploy AI. I ask: where in this organisation does taste actually live? Who is the person whose acceptance signals "yes, this is good"?

In a small business, taste lives in the founder. They picked the customers, the products, the tone, the standards. The taste is the business in a literal sense.

In a larger business, taste is meant to live in senior operators — the head of marketing, the head of product, the head of sales. Sometimes it does. Often, it has been diffused across committees, processes, and approval loops that no one person actually owns. When that happens, AI deployment becomes very hard, because the model has no single person to please.

The companies winning with AI tend to be the ones where someone — usually one person — has stepped forward and said: this is what good looks like, and I'll defend it. That person becomes the taste-keeper, and the rest of the organisation deploys AI in service of their standard.

Here's the part that's new and worth dwelling on.

For most of business history, taste was bundled with execution. The senior copywriter had taste and the ability to write the words. The senior designer had taste and the ability to push the pixels. You couldn't get the taste without paying for the execution, because the two were inseparable.

That bundle is broken. Execution is now near-free. Taste is now scarce in a way it never had to be before, because it can be expressed without the years of execution training that used to gate it. Someone with twenty years of senior judgment and zero ability to type can now direct an AI-led team that ships in days what used to take quarters.

The implication is brutal for some roles and amazing for others. If your value is execution at a level AI can match, you are about to be cheap. If your value is judgment about which execution to ship, you are about to be expensive.

Taste is the one thing in the AI economy that compounds. It is built by repeated exposure to the best of a category, paired with explicit work to articulate why it's the best.

If you run a business and you want to win in this environment, the time and money you'd previously have spent on hiring executors should go, in part, into developing taste — yours and your team's. Specifically:

• Spend an hour a week studying the very best work in your category. Not your competitors — the best. Write down what makes it good, in your own words.
• Make accept-or-reject decisions explicitly, and explain them out loud. Train the rest of the team to see what you see.
• Hire for taste, not for output volume. Outputs are now cheap. Judgment is not.
• Use AI to increase the pace of taste decisions — generate ten options, pick one, generate another ten — not to replace the deciding.

The teams that do this are leaving the rest of the market behind, quietly and quickly. The teams that don't are publishing more output than ever before, and getting less attention for it. AI has not destroyed standards. It has made them the only thing that matters.