App volume doubled this year. Usage grew 7%.

Across iOS, Android, and the Chrome Web Store, the number of new apps published each month has doubled over the past year, in some cases quadrupled. Over the same period, time spent in apps grew only 7%, and total category revenue moved just 2%, according to SensorTower's US market data.

A recent study looking at the same window checked an even more telling number: the share of apps hitting any sign of real traction at all, a threshold set at 10 ratings or 100 downloads on Android, 10 downloads on Chrome, has collapsed. This isn't just supply outrunning demand. Almost everything shipping right now doesn't even clear the lowest bar.

Two categories are the exception, and that's the interesting part

Looking at what's escaping the stagnation tells you more than the average does. Two categories are growing meaningfully faster than the rest: Productivity, driven by ChatGPT, Claude, Gemini, and Grok, and Dev Tools, driven by Replit.

That's not a coincidence. Neither one asks the user to learn a new product. Opening ChatGPT answers a need that's already there (ask a question, write something) with a tool people already understand. Using Replit does the same thing people already did, write code, just faster. Neither category is betting on product novelty. Both are betting on need familiarity.

💡 The contrast is worth sitting with: tools that give direct access to AI are capturing usage. Tools simply built with AI, mostly, aren't. The first solves a known problem faster. The second is often a problem nobody had yet, solved in record time.

Why faster code generation wasn't enough

This isn't a code-quality problem. A product that finds real usage has always needed the same three things, regardless of AI: someone with a real problem, a tool that solves it in a recognizable way, and a path for the right person to find that tool at the right moment. AI made the first piece, writing the code, nearly free. It changed nothing about the other two.

The source report itself asks the question that matters at the macro level, when this will actually show up in real economic numbers. Its answer: once the apps themselves get better, not before. And paid consumer AI adoption is still low, around 3% as of Q1 per Consumer Edge, with younger users adopting roughly 4x faster than older ones. Plenty of people are trying it. Not many, yet, are paying for it.

⚠️ A nuance worth keeping in mind: this market is still very young on this exact question. Some of what we're seeing today is likely amateur tinkering arriving ahead of demand catching up, not permanent proof that AI-generated code can never find real usage. This is a snapshot taken very early in the game, not the final score.

What this actually changes if you're building with AI

Three things are worth checking before your next sprint, not after:

1. Development speed has stopped being an advantage. Everyone has it now. If your plan rests on "we move faster than others because of AI," that's not a plan anymore, it's everyone's starting line.
2. Check that the need already exists, before checking that the product works. The two categories pulling ahead aren't teaching users a new reflex, they're speeding up a reflex that already existed. A product that first has to teach why it matters starts with a handicap that build speed doesn't make up for.
3. Have an answer on distribution before you have an answer on product. With more supply and flat demand, getting noticed gets harder, not easier. If "how will our first ten users find us" doesn't have a concrete answer, that's not a detail to sort out after launch.