When Machines Have Taste

There is a question that nobody in technology wants to ask out loud, because the answer is uncomfortable: now that AI can generate virtually anything, what should it generate? Not what can it make. What should it make. The distinction sounds philosophical. It is actually the most practical business question of our time.

We are living through the greatest explosion of creative output in human history. AI image generators produce millions of images daily. AI writing tools generate more text in an hour than the entire publishing industry produced in a year two decades ago. AI music tools, video tools, code tools, design tools. The volume is staggering. And most of it is mediocre.

That is not a criticism of the technology. The technology is extraordinary. It is a criticism of how we are using it. We have treated AI as a volume machine when we should be treating it as a taste machine. The difference between those two approaches will separate the companies that matter from the companies that drown in their own output.

The volume trap

The instinct, when handed a tool that can produce infinite output, is to produce infinite output. This is understandable and wrong. It is the content farm mentality applied to every creative discipline simultaneously.

Look at what happened when AI writing tools first went mainstream. Companies that had been publishing ten blog posts a month suddenly published a hundred. Marketing teams that had been sending one email a week sent five. Social media managers who had been posting once a day posted five times. The logic seemed sound: if the tool makes production cheaper, produce more.

The result was predictable. Engagement dropped. Quality dropped. Trust dropped. The companies that flooded the zone with AI-generated content did not win. They annoyed their audiences and diluted their brands. The volume machine ate itself.

Meanwhile, the companies that used AI differently are thriving. They did not use it to produce more. They used it to produce better. They used it to test more options and ship fewer. To explore more directions and commit to the strongest one. To draft faster and edit more carefully. They treated AI as a tool that increases the quality ceiling, not the quantity floor.

That is the taste difference. And it is becoming the defining competitive advantage in every industry that touches creative work.

Three companies that get it

Consider Stripe. A payments company, technically. But Stripe has taste that most design studios would envy. Their documentation is beautiful. Their developer experience is obsessively refined. They run Stripe Press, a publishing house that produces physical books about economics, technology, and progress. A payments company with a publishing imprint. That is taste in action.

When Stripe integrates AI into their products, they do it with the same restraint they apply to everything else. The AI features enhance the developer experience. They do not clutter it. They solve specific problems. They do not generate generic solutions. Stripe understands that taste means knowing when not to use the tool as much as knowing when to use it.

Consider Apple. The company has had access to AI capabilities for years. It could have shipped AI features across every product the moment the technology was ready. It did not. Apple waited. It refined. It chose specific use cases where AI genuinely improved the experience and ignored the rest. You can disagree with Apple's pace, but you cannot deny that the approach is taste-led. The restraint is the point.

Consider Notion. When Notion added AI to its product, it did something clever: it made the AI a collaborator, not a replacement. Notion AI helps you write better, not more. It summarises, restructures, and refines. It does not generate content from nothing and dump it into your workspace. The design choice reflects a belief about what AI should do, which is enhance human thinking rather than replace it. That belief is a taste decision.

All three companies share a common trait. They treat AI as an ingredient, not the meal. The technology is present but it is not the point. The point is the experience, the product, the thing the customer actually cares about. AI serves the vision. The vision does not serve AI.

Taste as quality control

Here is a framework that I find useful: think of taste as the quality control layer in any AI workflow.

In manufacturing, quality control is not optional. Every product that comes off the line gets inspected. Defects get caught. Standards get maintained. Nobody would run a factory without QC and expect to build a great brand.

But in the AI-powered creative economy, most companies are running without quality control. They are letting the machine produce and shipping whatever comes out. The volume is there. The consistency is not. And the audience can tell.

Taste is the QC layer. It is the human judgment that sits between the AI output and the published work. It asks: Is this good enough? Does this meet our standard? Does this feel like us? Does this add something or just add noise? Those are not questions the AI can answer about its own output. They require a human perspective, a set of standards, a point of view about what "good" means.

Companies that build this layer into their AI workflows will produce less but better. Companies that skip it will produce more but worse. Over time, the quality gap compounds. Brands with taste accumulate trust. Brands without it accumulate fatigue.

The design system analogy

The best analogy I have found for how taste should work with AI is the design system.

A design system is a set of constraints. It defines which colours, fonts, spacing, components, and patterns a team can use. It limits choice. And by limiting choice, it ensures consistency. Every page, every screen, every interaction feels like it belongs to the same product. The constraints are not restrictions. They are taste, codified.

AI workflows need the equivalent. Not a free-for-all where anyone can generate anything and ship it. A set of principles, standards, and guardrails that define what the AI should and should not produce. A taste system.

What does our brand sound like? What topics do we cover and what do we ignore? What level of quality constitutes "good enough to publish"? What is the ratio of AI-generated to human-refined in our final output? These are taste questions. And they need answers before the AI starts generating.

The companies that build taste systems will scale their AI usage without diluting their brands. The companies that do not will discover that AI-at-scale without taste-at-scale produces exactly what you would expect: a lot of forgettable work.

The uncomfortable question

All of this leads to an uncomfortable question for leaders and founders: does your company have taste?

Not "does your company have a brand guidelines document." Not "does your company have a design team." The question is more fundamental than that. Does your company have a genuine point of view about what is good and what is not? Can your team articulate the difference between "on brand" and "off brand" without checking a PDF? Do the people making daily decisions about what to publish, ship, and promote have a shared instinct for quality?

If the answer is yes, AI will make your company dramatically more powerful. The taste is the constraint. The AI is the accelerant. Together they produce more of what is genuinely good.

If the answer is no, AI will expose the gap. More output with no taste filter means more mediocrity at scale. And audiences are increasingly skilled at detecting the difference between something made with care and something generated without it.

The machines can have whatever capabilities we give them. The question is whether we have the taste to direct those capabilities toward things worth making. That is not a technology problem. It is a leadership problem. And the companies that solve it will define what the next era of creative work looks like.

Next in this series: The Curation Premium, on why selection is the new competitive advantage.