A friend called me last week with a strange problem. He's built several software products — real ones, live, working. He didn't hire a developer. He didn't write a business plan. He opened an AI subscription, described what he wanted, and a few hours later he had something that ran. He works in a field that has nothing to do with technology.
His question wasn't about bugs or hosting. It was simpler than that, and harder: he doesn't know how to sell any of it, or who would buy it.
My answer wasn't comforting, but it was honest. Great, I told him — you saved the money it would have cost to hire a developer. Now spend that money on marketing. That's where the actual problem was hiding the whole time.
The filter that used to exist
Tell someone you're a software developer at a dinner party and watch what happens. Within five minutes they'll pitch you an idea. Everyone has one. Ideas are free, which is exactly why there's always an overwhelming number of them floating around.
What used to happen next was brutal, and it happened quietly. A small percentage of ideas made it into actual development, because development cost real money and required finding a real developer willing to take the risk. A smaller percentage of those became a finished product. And a smaller percentage of finished products ever reached a market and found people willing to use them.
That chain of filters wasn't designed by anyone. It was just friction — the cost of a developer's time, the cost of getting something built, the cost of being wrong. Friction is annoying, but it does a job. It kills weak ideas before they waste real resources.
What AI actually removed
AI didn't make people more creative. The overwhelming supply of ideas was never the problem — people have always had them. What AI removed is the friction between having an idea and holding something that runs.
Open a chat window, describe what you want, and within hours you have a product. Set aside whether it's built well, whether it's secure, whether the architecture will survive its first real user. Set that aside entirely — it still runs, and to someone who's never built software before, that's indistinguishable from success.
So we've moved from one kind of overwhelm to another. It used to be an overwhelming number of ideas, most of which died quietly before costing anyone real money. Now it's an overwhelming number of finished products, most of which nobody asked for and nobody wants — except the filtering didn't happen before they existed. It has to happen after, in public, the hard way.
The bottleneck moved. It didn't disappear.
My friend's situation makes the point better than any argument I could construct. He built several products entirely on his own, for a fraction of what development used to cost. By the old math, he should be ahead. He isn't — because the part of the process that actually determines whether software succeeds was never the building.
The hard part of software was never writing it. It's making someone want it who didn't ask for it.
Distribution, positioning, pricing, trust, the unglamorous work of finding ten people willing to pay before you find ten thousand — none of that got easier. If anything it got harder, because now it's competing against a flood of other AI-built products making the exact same noise in the exact same channels.
This isn't an argument against building fast. We're a custom software company and we've adopted AI tools ourselves — they're genuinely useful for a lot of what we do. The argument is about sequencing. Building fast doesn't help you if you're building the wrong thing fast.
What actually filters an idea
The filtering process didn't go away. It moved downstream, from "can this be built" to "does anyone want it." Since AI won't ask you these questions before you start typing, you have to ask them yourself:
- Who exactly is the customer? Not "small businesses." A name, a job title, a specific frustration they already have.
- Will they actually pay? Not "would find it useful" — will they take money out of their pocket for it.
- How will they find you? A channel you already have access to, not one you're hoping to build from nothing.
- What does it cost to acquire one customer? If you don't know, you don't have a business yet. You have a working demo.
- Have you talked to ten of them before building anything? Not surveyed. Talked to. If the answer is no, that's the actual first step — not the next feature.
None of these questions require a developer, an AI subscription, or a single line of code. They're cheaper than building the product and they're the reason most products that skip them end up exactly where my friend did — finished, functional, and unsold.
