I was at Panathenea in Greece at the end of May, and everyone kept asking me the same thing: what is product engineering? CAIOs asked it. Founders asked it. Plain engineers asked it. My answer kept getting shorter, and I think that's the point.
The same people, or people just like them, ask me a second question. Usually right after. "If AI can build everything, where do humans still matter?" They think it's a different question. It isn't. It's the same one wearing a different shirt.
So let me answer both. And I'm going to do it with the help of a 16th-century Japanese aesthetic concept, because it turns out it does more work here than any framework I could draw on a whiteboard.
This isn't a flex. It's just the clearest way I've found to explain what I actually do all day.
So what is it, actually
The clean answer is that product engineering is the intersection of product, engineering, and taste. And that's genuinely true, I won't pretend otherwise. But "intersection" makes it sound like a Venn diagram on a slide, three circles politely overlapping. The reality is messier and more interesting than that.
It's owning the outcome end to end. Framing the problem. Judging what's actually feasible. Having taste about how it gets built. Shipping it fast. And knowing, in your gut, when "good enough" beats "perfect" and when it absolutely doesn't. Product and engineering are the two ingredients. Taste is what turns them into something that's actually worth shipping.
The old world had a loop. Product writes the spec. Engineering implements the spec. It goes back and forth across a wall, with handoffs and tickets and meetings about the handoffs and tickets. That loop made sense when writing the code was the expensive part.
Writing the code is not the expensive part anymore.
So the loop collapses. Product engineering is what's left when you remove the wall. One person, or one tight team, holding the whole thing: the why, the what, and the how, all at once. I lived this at Native Teams before we had a clean name for it. Navigating PMF ambiguity with no clear spec to hand anyone. Migrating a whole stack off Vue to React. Building a design system. Then turning the entire team AI-native. None of that work fit neatly on one side of the wall. The wall was the problem.
The bottleneck moved, and most people missed it
I haven't written code in about a year. The work didn't vanish. It moved up a floor.
What I do now is closer to directing. I describe what I want. I set constraints. I review what comes back. I course-correct. I ship. The models do the writing. I do the thinking. My agents have written something north of 346,000 lines in the last year. The feedback loop that used to take days now takes minutes.
Here's the part people miss. When execution gets cheap, execution stops being the scarce thing. You can generate ten implementations of a feature before lunch. The scarce thing is knowing which one of those ten is right. Not which one compiles. Not which one passes the tests. Which one is right.
That's not an engineering skill in the old sense. That's taste. And it's the whole job now.
Taste is the moat. And it's earned, not born.
AI can give you a thousand right answers. Taste is knowing which one isn't just correct, but right.
Because "right" isn't always optimal. It isn't always the most polished, the most complete, or the most statistically likely continuation of the prompt. Right is contextual. It depends on who you're building for, what you're willing to cut, and what the thing is supposed to feel like when someone actually uses it. A model optimizes toward the average of everything it has seen. The average is rarely the right answer for your specific problem.
And taste is not innate genius. This is the part I want people to hear. Taste is acquired. You earn it through exposure, through shipping things that flopped, through cultural context, through working inside real constraints, through time. You cannot prompt your way into it on day one. There's no system message that gives you ten years of "I've seen this fail before."
This is exactly why I hire for judgment, not just skill. The best engineers on my team are not the fastest coders. Speed is the machine's job now. The best ones are the ones who look at a perfectly reasonable-looking AI output and go "no, that's wrong" and they're right, and they can't fully explain why except that they've felt it before. They catch the subtle wrongness. They redirect the agent. That instinct is the asset.
I'll be honest about the shelf life. This moat is real but I don't think it's permanent. Models will get better at pattern-matching what good taste looks like. They already are. But pattern-matching the output of taste is not the same as the acquisition of it, and acquisition still runs at human speed. For now, that gap is where we live.
Wabi-sabi, or why perfect feels hollow
Here's where a 16th-century Japanese idea earns its place. Wabi-sabi (侘寂) is the appreciation of beauty in the imperfect, the impermanent, the incomplete. The hand-thrown bowl with the uneven lip. The crack repaired with gold instead of hidden. The mark of the maker, left visible on purpose.
AI defaults to the exact opposite. It smooths. It centers. It resolves. It produces the median of everything it has ever seen, and the median is, by definition, the most average version of good. Technically flawless and somehow hollow. You've felt it. The copy that reads fine but says nothing. The landing page that's clean but forgettable. The thing with no fingerprints on it.
What makes work actually resonate is almost always the human signature in it. The deliberate rough edge. The opinionated choice no average would ever land on. The part that's a little "wrong" in exactly the right way. That's taste, and it's the most wabi-sabi thing there is: understanding that perfect and good are not the same word.
This is why "ship the imperfect version" isn't laziness, it's a position. Not everything needs to be artisanal hand-crafted code. Some things just need to work, and the speed of learning beats the polish. But the things that are meant to be felt, the product's voice, the moments a user actually notices, the decisions that give a thing character, those need a human fingerprint. Because a fingerprint is the one thing the median can't fake.
None of this means AI is the enemy
If this reads like I'm romanticizing the human hand, let me correct that. I'm not.
AI is the single biggest enablement of human capability I've seen in my career. Not a replacement. Not a savior. An enablement. It strips away the friction so you can operate a whole level of abstraction higher than before. I build side projects in weekends that would have taken months. We've shipped mobile apps at Native Teams at a pace we genuinely did not think was possible a year ago. That's not theory. That shipped.
And the gap between the people using this and the people watching is compounding every week. The teams building AI muscle memory now are pulling away, and you can't download that muscle memory overnight. Small fast teams are starting to eat slow orgs that are still scheduling the meeting about whether to allow the tools.
So no, the answer is not "humans good, machines bad." The answer is a division of labor. Humans do what humans are still best at: taste, framing, judgment, knowing when to stop. Agents do what they're best at: execution, volume, iteration, never getting tired. You point, they build, you decide. That loop is faster and better than either side alone.
The director with taste
So here's the shortest answer I've got.
Product engineering is the intersection of product, engineering, and taste, applied at the speed of agents. It's the discipline of pointing acquired judgment at AI-enabled execution and shipping the result. Wabi-sabi is one expression of that taste, the part that knows perfect and good aren't the same word, and that the human fingerprint is what makes work land. AI is the engine. Taste is the steering. The moat was never typing the code. The moat is knowing what to build, what to reject, and what to ship before it's perfect.
The engineer who thrives from here isn't a typist. The machine types now. And they're not some detached philosopher either. They're a director with taste. They know the difference between right and merely correct, and they earned that difference the slow way, which is still the only way.
That's what I tell the people who ask now. It used to take me a paragraph. Now it takes a sentence.
The machine builds. You decide. That's the job.
- D