Anything you can describe, you can have built by Friday. Feature, product, second business, the thing you have wanted for three years. Describe it, wait, it exists. This is being sold to you as the best day of your working life, and for the part where you used to be stuck, it is. For the part where you actually decide, it just moved the wall and turned off the lights.
For all of recorded work, the hard part was making the thing. You wanted to build something and the world sent back an estimate: six weeks, a skill you did not have, a budget you could not get, a team you had to hire and manage and keep. Most of what you could imagine died at that gate, quietly, before you ever weighed it. It felt like restraint. It was an invoice. The cost did your choosing for you and never told you it was choosing, and you walked around feeling like a person with good instincts when you were a person with a small menu.
The tools took the invoice away. Not lowered it. Removed it. And the moment building costs nothing, every question that used to sound like planning turns out to have been standing in for a bill. Can we build it. Yes. Can we build it in time. Yes. Do we have the skills. The model does. Every filter you were quietly leaning on ran on scarcity, and the scarcity is gone, so the filters return the same answer for the brilliant idea and the terrible one, which is yes, which is worth exactly nothing.
So the thing everyone is celebrating is the thing that broke your decision-making, and nobody sent that memo. When making was expensive, being able to make something was the whole game, and the person who could build won. Now everyone can build. Capability stopped being a sorting mechanism the instant it stopped being rare. It is table stakes. It distinguishes no one, which means the entire weight of the decision has slid onto the one part the tools do not touch: choosing what, out of everything now buildable, is actually worth the building.
That is a different skill, and it is the one nobody trained for, because for most of history you did not have to. The menu was short enough that picking from it felt like judgment. You never had to look at everything and decide, because everything was never on offer. Now it is. The infinite menu is not a gift. It is a test you were not studying for, and it grades on a curve where everyone else got the same tools you did.
People hear "AI can build anything" and reach for taste. The eye. The exquisite sense for what is good. It is the flattering answer, so it sells, and it is the wrong department entirely, because taste rates one thing at a time and a bottomless menu is not a rating problem. Rate every option and they all come back good, because a tool that says yes to everything is a mirror, not a filter. The work is comparative, and it is downstream, and it is graded by reality months after you commit, none of which is what taste does. Taste and the scarcer thing under it, discernment, are a whole essay of their own, so I will hand that argument to where it lives and not relitigate it here. The one line to carry forward is smaller: when capacity is free, the bottleneck is judgment, and judgment is about what to build.
Which sounds like advice and is not, yet, because "have better judgment" is the same empty instruction as "have better taste" until someone breaks it into moves you can actually run. So the cluster breaks it into four.
The first is the one you hit first, before anything gets built: selection. You are standing at the bottomless menu and every third thing on it would actually work, and capable people do not freeze at bad menus, they freeze at good ones. The move is to put back a constraint the tools took off, name the one resource you are short of, and rank the options against each other instead of rating them one by one. That is its own piece.
The second sits next to it and points the other way: the no-build. Cheap building did not lower your risk, it flipped it. The danger used to be building too little because you could not afford your good ideas. Now it is building too much, because the build was never where the money went, the maintenance tail was, and every shipped thing is a standing liability you took on for a one-time hit of shipping. Deciding what not to build is the strongest competitor you have, and it has its own piece, because the no-build is invisible and nobody screenshots the feature you declined to ship.
The third is subtler and it is where the machine actively works against you: the question. A model gives you a beautiful answer to whatever you asked, including when what you asked was the wrong thing, and the polish of the answer is exactly what stops anyone reopening the question underneath it. Shorten the onboarding, it will, brilliantly, without once asking whether anyone should have to onboard at all. That failure has its own piece too.
The fourth is the one people assume they already have and mostly do not: the judgment itself, and where it comes from. Not from years. Twenty years in a trade does not settle into judgment on its own, it laminates the same confidence you started with under two decades of stories, because a decision that gets graded late and tangled with everything else is a terrible teacher. Judgment is built, not accumulated, and building it is a practice with steps, and that is the last piece.
So the honest counter, before you leave, because it is the best live argument against all of this. Bias to action. The whole point of cheap building, the argument goes, is that a wrong build stopped being expensive, so stop deliberating, build the thing in an afternoon, let reality kill the losers, and quit strangling your weirdest ideas in a should-we meeting. For a real class of builds this wins outright, the throwaway things, the reversible ones, the ones nobody comes to depend on. Build those, ship small, let reality grade them, and skip everything I just said. The counter only breaks on the builds that look cheap because the build step is cheap and are not reversible at all, because the second they ship they collect users and obligations and a claim on every future decision. Those are the ones the judgment is for. Sorting which is which is the practice, and it is the whole cluster.
None of this hands you the constraint back. That is the part worth being clear-eyed about. The tools removed the one thing that used to make the choice for you, and they are not going to replace it with a new one, because a tool that says yes to everything cannot also be the thing that says no. The no is yours now. It was always going to end up yours. It just used to arrive disguised as a budget.
Common questions
- What should I build now that AI can build anything?
- Build what survives a judgment about worth, not a check on capability. When building is free, "can we build it" is true of everything and decides nothing, so the useful question becomes which few things, out of everything now buildable, are worth the attention, the maintenance, and the commitment. Choosing is the work; the tools removed the constraint that used to do the choosing for you, and did not replace it.
- Why is it harder to decide what to build now that AI makes building cheap?
- Because the cost of building used to filter your options before you weighed them, so you only ever chose among the few things worth affording. Cheap building deletes that filter. The menu fills with good options, not bad ones, and every capability question now returns yes, so the full weight of the decision lands on judgment about worth, which most people never had to practise.
- Is taste the scarce skill now that AI can build anything?
- No. Taste rates one thing at a time, and when every option comes back good, rating cannot cut the set. The scarce skill is judgment about what to build: comparing options against a real constraint, pricing the whole life of a thing rather than its build, auditing the question you are answering, and grading your own calls against what actually happens.