Hire for taste, not craft: What AI actually replaces on a product team

The most skilled person on your team may no longer be the most valuable one. Craft was the differentiator. The designer who could execute a pixel-perfect prototype, the engineer who wrote clean code on the first pass, the copywriter who nailed tone without a brief. Those people were worth their weight in equity.

But something has shifted beneath every hiring conversation in product. AI can now do craft. Not all of it. Not perfectly. Well enough, though, and fast enough, that raw execution skill is no longer the scarce resource it was. The question product leaders should be asking is not who can build the best version of this. It is who can tell me whether we should build it at all.

The craft floor

I have started calling it the craft floor. The craft floor is the minimum level of execution quality that AI tools have made universally accessible. Two years ago, producing a high-fidelity prototype required a senior designer with years of practice. Today, a junior designer with Figma and an AI assistant can produce something that looks almost indistinguishable. Writing clean front-end code once required specific expertise. Today, an AI coding tool can generate it from a description.

Here is what the craft floor does to hiring. It compresses the value of execution skill. When everyone can produce work that clears the quality bar, the bar stops being a differentiator. It becomes the floor, the minimum. The thing that gets you into the room but no longer determines who wins.

And what wins, increasingly, is something harder to test for, harder to teach, and impossible to automate. Taste.

Two designers, one quarter

At Adobe, we hired two designers in the same quarter. One of them, I will call her Priya, had exceptional craft skills. Her prototypes were flawless. Her visual design was precise to the pixel. She could take a brief and produce something polished faster than anyone I had worked with. In a portfolio review, she was the obvious hire.

The other designer, I will call her Meera, had less polish. Her prototypes were rougher. Her visual sensibility was good but not remarkable. In her first month, she killed a feature. Not dramatically. She just looked at the problem the team was solving, talked to three customers, and said this is not the right thing to build. She could not articulate exactly how she knew. She was right. The data confirmed it six weeks later.

Within a year, Meera had measurably more impact. Not because she was a better designer. Because she kept the team from building three features that would have failed. She saved months of engineering time and opportunity cost. Priya built beautiful things. Meera built the right things. The gap between those two contributions is what I have come to call the taste premium.

The taste premium is the disproportionate value created by people who can distinguish between work that is good and work that matters. Craft is the ability to execute well. Taste is the ability to know what is worth executing. And in a world where AI has raised the craft floor for everyone, the taste premium is widening fast.

The intern who could execute anything

I have been mentoring a junior designer who started as an intern. She picked up AI tools faster than anyone I have seen. She could generate interfaces, write copy, build interactive prototypes, all at a pace that would have been senior-level output three years ago. Her craft, amplified by AI, was genuinely impressive.

But she built a polished, beautiful interface for a feature that solved no real problem. The screens looked professional. The interactions were smooth. And the entire thing was a solution searching for a problem that did not exist. When I asked her why she had built it, she said the brief mentioned it as a possibility and the AI tools made it easy to explore. She could execute anything. She could not tell when something was wrong.

That is the craft floor in action. The tools gave her the ability to produce far beyond her experience. They could not give her the instinct to stop and ask whether the output was worth producing. The craft was there. The taste was not.

Two guitarists can play the same piece. One hits every note with technical precision, every chord clean, every transition smooth. The other leaves notes out. Pauses where the sheet music says to play. Holds a silence one beat longer than expected. The first guitarist is skilled. The second one is an artist. The difference is not what they can play. It is what they choose not to play.

Craft is what AI can replicate. Taste is what it cannot.

Product teams are full of people who can play every note. AI has made that easier than ever. But the people who know which notes to leave out, which features to kill, which elegant solution to reject because it solves the wrong problem, those people are the most valuable on any team. And most hiring processes are not selecting for them.

What hiring for taste actually looks like

But if taste is the new premium, most hiring processes are still testing for craft. Portfolio reviews evaluate visual quality. Take-home exercises measure execution speed. Technical interviews test implementation skill. All of these assess whether someone can build something well. None of them assess whether someone can tell you what not to build.

Hiring for taste means asking different questions. Not "show me your best work" but "show me something you decided not to ship, and why." Not "how would you design this feature" but "how would you decide whether this feature should exist." Not "walk me through your process" but "tell me about a time your process told you to stop."

But here is the harder truth. Taste is difficult to interview for because it looks like confidence without evidence. When Meera said the feature was wrong in her first month at Adobe, she did not have a spreadsheet to support it. She had a feeling informed by years of watching users, an instinct she could not fully explain. That kind of judgment makes hiring managers nervous. It is easier to evaluate a pixel-perfect prototype than a well-reasoned no.

But the well-reasoned no is worth more now than it has ever been. Because AI can produce the prototype in minutes. Nobody can automate the judgment that says this prototype solves the wrong problem.

The split that is coming

The craft floor will keep rising. AI tools will get better at generating code, design, copy, and research summaries. Every execution task that can be described can eventually be automated. But this is not a threat to product people. It is a filter. It is separating the people whose value was in execution from the people whose value is in judgment. And judgment, the instinct for what is right and worth shipping, is the thing no model is coming for.

The best product people I have worked with, across petroleum, aviation, SaaS, and retail, all shared one trait. They knew when to stop. They could look at a well-crafted solution and say this is not it, without a metric to prove them right in the moment. They were usually proven right later. That is taste. It is pattern recognition refined by years of watching things succeed and fail. It is the residue of experience that no training data can replicate.

The teams that hire for it will build things that matter. The ones still hiring for craft will build things that look good. Those are not the same outcome.