“There’s an app for that” was a great advertising line, and for most of the smartphone era, it’s largely been true. Since anyone can build an app for a phone, the sheer variety of niche solutions exploded. But is an app store full of near-perfect solutions really the future?
The app store solved yesterday’s problem
It’s easy to forget that the original iPhone launched without an app store and no third-party software at all. Feature phones that predated smartphones let you easily download and run Java apps. When Apple and Google finally nailed the app store concept, they centralized everything — no more visiting sketchy websites to find apps, no more compatibility worries, and a dramatically reduced risk of malware.
This all made sense because coding an app was hard. Reducing friction between developers and users was essential. But with the rise of AI coding tools, that may no longer be the case.
Most people don’t actually want apps
Apps have always been a means to an end. Nobody wakes up thinking “I’d like to install another app.” We go looking for apps when we have a specific problem to solve. Someone who downloads a habit tracker isn’t buying a habit tracker — they’re buying a way to build better habits. Someone who installs a note-taking app doesn’t want software; they want a place to organize their thoughts.
Here’s why even great, simple apps don’t stay that way: feature creep. Users keep asking for more, and developers feel obliged to deliver. That’s how we end up with bloated, unfocused apps that used to be brilliant.
The best software might be software nobody else uses
Now imagine this: instead of browsing an app store for something close to what you need, an AI coding model creates a simple app or function that solves your exact problem. An app that’s private to only you. No one else can download it, use it, or find it.
Tools like Claude Code may need professional supervision for complex apps, but they’re already good enough to build small applets that serve a few functions. If the push toward agentic operating systems yields results, this is something we’ll see. The agents trying to serve your needs within the OS could simply write the function you need when you need it. It still feels like a wild idea, but the building blocks are already here: capable coding LLMs, agent frameworks, and hardware that runs models locally.
A future without redesigns, subscriptions, and feature creep
If this era of personalized AI-coded applets arrives, it changes a lot about how we use our devices. Today’s apps aren’t created purely to serve users — they’re built for engagement, subscriptions, revenue, and long-term business growth.
Nobody likes unwanted redesigns, apps that get slower over time, or previously free features suddenly paywalled. With an on-device agentic coding solution that adapts to your specific needs, you wouldn’t have to settle for lowest-common-denominator software.
Google and Microsoft are already heading this direction. Google’s “Create My Widget” feature uses generative UI to build personalized widgets with Gemini. Microsoft is trying to make Windows an “agentic” OS, and while Copilot hasn’t worked out the way the company hoped, that doesn’t mean they’ll stop trying. Don’t be surprised if we reach a point in the near future where a coding bot lives on your device and just adds the features you need without involving an app store or a human developer.
