Last year, shipping an iOS app meant weeks of Swift tutorials or paying a freelance developer $80-150/hour. Today, you can describe the app you want and have a working Xcode project in a few hours.
One developer recently documented this experience across 9 projects: 4 complete apps on the App Store - SwiftUI interfaces, StoreKit in-app purchases, widgets, Live Activities - with 5 more in active development. Time to write the product requirements documents: longer than the time to get working builds. Total revenue across all 4 shipped apps: $0. Total active users: roughly two people, one of whom probably downloaded by accident.
This is not a story about AI tools falling short. Claude built every app exactly as described. The code compiled, the apps passed App Store review, and the features worked. The constraint wasn't technical execution. It was figuring out whether anyone needed these apps in the first place - a question that never got answered before building started.
The Barrier That Moved
AI coding tools eliminated technical execution as the bottleneck. You don't need to know Swift to produce working SwiftUI code. You don't need to understand StoreKit to implement in-app purchases. Claude Code and similar tools handle the implementation; you decide what to build.
What they can't do is validate demand. They can't tell you whether 40 apps already cover this niche, whether the target audience is large enough to generate any revenue, or whether anyone will pay for what you've built. That work needs to happen before a single line of code gets written - and when writing code is fast and nearly free, it's easy to skip it entirely.
The practical fix is straightforward: when building is no longer the expensive part, the PRD (the document where you work out exactly who the app is for, what problem it solves, and why they'd pay for it) becomes the real product. Five 20-minute conversations with potential users before opening Xcode would have outperformed five more Claude-built apps with no audience.
Not Just Apps
The same dynamic shows up wherever AI tools have reduced execution costs. AI writing tools make content fast to produce; they don't make it rank or get read. AI image tools make visuals easy to create; they don't make them convert. The execution barrier has dropped significantly across almost every creative and technical category. The strategic question - what to build, for whom, and why - hasn't gotten easier at all.
Demand validation (talking to potential users, checking search volume, reviewing what's already available in the category) is now the highest-leverage part of any build process. The actual development has become almost incidental.
The AI tools delivered exactly what they were built to do. The apps just didn't need to exist.