AI Made Building Free. Taste Is the New Bottleneck.

In February 2025, Andrej Karpathy — who ran AI at Tesla and was a founding member of OpenAI — described a new way of making software. He called it "vibe coding": you "fully give in to the vibes, embrace exponentials, and forget that the code even exists." To prove it, he built a working iPhone app in about an hour, in a language he didn't know.

Sit with what that means. The thing my whole industry treated as the hard, expensive, gate-kept part — turning an idea into a working thing — just collapsed to an afternoon and a conversation. I feel it in my own work: I've coded a live airline-operations app with AI in five days, taught myself an ERP and built it on the ground in four weeks. Building used to be the wall. It isn't anymore.

So here's the question that should be keeping product people up at night: if anyone can build anything, why is so much of what gets built still worthless?

You just felt it. The model handed you everything; the work was deciding what to throw away — and notice, nothing you added improved it, removing did. That judgment, about what actually matters to a real person, has a name, and it's about to become the most valuable thing you own.

When one input goes free, value moves to the scarce one

This is an old economic pattern, playing out very fast. When one ingredient of a thing becomes abundant and cheap, the value migrates to whatever is still scarce. Generation just became abundant. So what's scarce now?

Not the building. Not the model — everyone has the same ones, and they improve every month. What's scarce is taste: the judgment to know what's worth building, what "good" actually looks like, and what to throw away. It's the one thing AI can't hand you, because it's the thing that decides what to ask the AI for in the first place.

And its absence has a name now too. When building was hard, bad ideas died of friction — they were too expensive to make, so most never got built. Take the friction away and they all get built, beautifully. The result is the flood everyone has started calling slop: confident, polished, professional-looking, and pointless. Slop isn't a model problem. It's a taste vacuum, filled at machine speed.

Taste isn't vibes. It's three hard judgments.

"Taste" sounds soft, like a matter of opinion. It isn't. From someone who ships, it's three concrete things AI doesn't do for you:

• Knowing what's worth building. The hardest question was never "can we build it?" — it was "should this exist, for whom, and why?" That's a question about a real human in a real situation, and you can only answer it by understanding them better than the next person does. AI will build the wrong thing flawlessly and never once ask if it should.
• Knowing what "good" is. AI gives you something that looks 90% done in minutes — and the last 10%, the gap between "demos well" and "actually holds up," is invisible unless you have a standard to measure against. Taste is that standard. Without it you ship the 90% and call it finished, because you literally can't see what's missing.
• Knowing what to cut. When adding things is free, everything gets added. The dashboard grows twelve charts; the app grows nine features; the flow grows six steps. Taste is mostly subtraction — the discipline to remove what's merely possible so the one thing that matters can be seen. The flood is an addition problem, and only taste solves it.

Notice these are all the same muscle: understanding the person you're building for well enough to judge, on their behalf, what's worth their attention. That's not a new skill. It's the oldest one in design — it was just hidden behind the months of building. Now the building is gone, and it's standing in the open, exposed as the entire job.

Why this is good news (if you have it)

For years, taste was undervalued because building was the bottleneck — the person who could ship was worth more than the person who knew what to ship, because shipping was the rare thing. That's inverted. Speed is table stakes now; everyone has it. Speed without judgment just means you build the wrong thing faster — and now everyone can.

So the moat is no longer "we can build it." Everyone can build it. The moat is the judgment to aim all that cheap building at the few things that matter, and the standard to know when they're actually good. In a world of infinite supply, the scarce — and therefore valuable — skill is discernment. The curator beats the generator. (It's the same muscle as knowing when to reach for AI at all.)

Taste is trainable — start now

The good news: taste isn't a gift you're born with, it's reps. You build it the way you build any judgment — by getting close to the people you serve, seeing a great deal of work both good and bad, and developing a strong, fast "no." A few starting moves:

• Get proximate. Taste is downstream of understanding. Watch real people use the real thing — the closer you are to the user, the sharper your sense of what's worth building becomes. You cannot have taste about someone you've never met.
• Build a standard. Study work that's genuinely good in your domain until you can feel the difference between fine and excellent. You can't aim at "good" if you've never calibrated what good is.
• Practise subtraction. For everything AI hands you, ask not "what could I add?" but "what can I remove and lose nothing?" Make cutting a reflex.

The reframe is simple, and it changes what you should be investing in. The question used to be "can you build it?" For the first time in my career, the honest answer is almost always yes — so it has stopped being interesting. The only question left is the one AI can't answer for you: should this exist, and is it any good? That's taste. In an age of infinite generation, it's the last scarce thing — and the best investment you can make.

Trying to decide if something's worth building? That's a taste question — and The Bet is built to help you make it honestly, before you spend the afternoon (or the quarter) building it. Or let's talk through what's worth your team's attention.