AI is the End of Apps ⭐

We’re so focused on how AI is changing the apps we know and love that we’ve lost sight of what’s really happening here. But this isn’t about changing apps, it’s about changing everything. And when it comes to apps, there’s only one endgame here. Apps are going away. We’re looking at the death of apps as we know them.

If you pay attention to Cory Doctorow, the person who invented the perfect term enshittification, and you should, you know that one of the themes he brings up repeatedly is that the only computer we know how to build is what he calls a Turing-complete universal von Neumann machine. There’s a lot that goes into that name, but to overly simplify it, what he means is that all general-purpose computers can run any program (app) that can be written for any general-purpose computer regardless of architecture or operating system. He brings this up in the context of enshittification, of course, as platform makers specifically and artificially limit which programs (apps) can run on their computers. The most extreme version being mobile platforms (iOS and Android), which are controlled by their makers.

When you think about the personal computing platforms you use, there is likely so combination of classic desktop platforms (PCs, Macs) and mobile platforms (iOS, Android). One of the many ways in which these platforms can be compared or even identified is through their apps, these familiar interfaces we use each day. And the ways we use these apps today are likewise familiar, with mouse/touchpad and keyboard being most common on the desktop and multitouch being most common on mobile. These are direct interactions: You launch an app and you use that app by directly interacting with it.

Back in August, Windows lead Pavan Davuluri appeared in an in-house Microsoft video about his team’s vision for the future of Windows, and the focus was largely on how AI will change the interaction experience in this desktop platform. And in his view, the Windows of five years from now will be increasingly “authentic” (whatever that means) and “multimodal,” an AI term that I believe he was purposefully contorting to use in an interaction sense. That is, in addition to the mouse/touchpad, keyboard, pen, and multitouch interactions we have today, we will have “voice and vision.” These are interaction types we see today in Copilot for Windows 11, albeit in early and ever-evolving forms. And Copilot, you may recall, was once described by Microsoft CEO Satya Nadella as the new Start menu, the “orchestrator of all your app experiences.”

Windows enthusiasts/power users of a certain age, meaning roughly my age, reacted in horror to both of these revelations, made two years apart. That’s what we do. But with the context of history, it’s useful to understand what each meant by these comments.

I can’t say that I knew what to think of Nadella’s Start comments at the time—remember, this was just as Copilot was messily rolling out before the official release of Windows 11 version 23H2, ahead of many changes to come, and before Copilot+ PCs and that damned Copilot key that’s everywhere now—but it most likely came off as an attempt to put Copilot in context to the billion-plus people who had growth used to Start as a primary Windows interaction point.

Davuluri’s comments were more easily explained, and not just because of the two years of updates we’d gotten to Copilot and Windows 11 since the Nadella quote. It’s clear that natural language interactions—both spoken and listened to—will be a key interaction type going forward on Windows and all personal computing platforms. And we know that the underlying AI interacting with what we see—out in the world with a mobile device, perhaps, and on-screen more typically in Windows—is key as well. AI is going to interact with us, and it’s going to interact with the broader world around us.

Getting from here (the present) to there (the future) will require multiple steps and the evolution of all kinds of things, across hardware, software, and services. And some of this will differ according to the device(s) we’re using. But keeping this to Windows for now, you can see how things change. Each week, it seems, Microsoft adds individual features to Copilot in Windows 11 and to the other increasingly AI-powered apps and services we use. These are overt and they behave much like news features in the system and in apps have always worked. We see them, we interact with them, and we use them using the same PC interface we’ve long had.

Less visible but no less ongoing are the changes happening under the covers, so to speak, at what I think of as the services level. This is where existing and new apps become programmatic, exposing individual features as services that can be accessed externally. In the past and using previous app-to-app sharing models like Copy/Cut/Paste, Dynamic Data Exchange (DDE), OLE (Object Linking and Embedding), and so on, these programmatic capabilities worked at the OS and then network level before flaming out at the Internet level when Microsoft tried and failed with ActiveX. But those and later service-like interactions are a good grounding, so to speak, in what’s happening to apps now.

If you’re familiar with some of the changes in Windows 11 in recent months, some of which are exclusive to Copilot+ PCs, you know that Microsoft is adding something called App Actions to the system and that the early interaction point here is Click to Do, which uses vision (Copilot Vision) to see what’s on-screen, highlight the text and images it sees, and then provide a series of actions, some system-based and some app-based, that you can perform on those things.

You might also know that these actions are being exposed directly in the file system and that you can do things like right-click a document or image file and see a growing series of things—actions—we can perform on them when they’re right-clicked. These actions are making the originally sparse Windows 11 context menus quite long in some cases. When I right-click an image file, for example, I see AI actions like Visual Search with Bing, Blur background with photos, Erase objects with photos, and Remove background with Paint in addition to choices like Ask Copilot, Edit with Clipchamp, Edit with Paint, and many, many others. And these are in addition to familiar options like Open with and whatever else.

In those options you can see apps become programmatic. It’s early days, but if you right-click an image and choose “AI actions > Remove background with Paint,” then the expected happens: Paint opens with that image loaded and its background removed. This is arguably more efficient than running Paint, opening that same image, and then finding the “Remove background” button and using it. But it gets you to the same place.

Where things get interesting is when we don’t need to open Paint anymore. If you use that same right-click interaction and trust this to actually work, you may want it to just remove the background. Of course, no one wants that: You might want to fine-tune the background removal and/or save the result as a new file. But those are capabilities of Paint, too. And they can and will be made programmatic. And that means that Windows could orchestrate all those actions in a single action.

That’s useful and it’s happening. But this gets even more interesting—or, if you’re a power user of a certain age, concerning—when you introduce natural language interaction to this workflow. You’re looking at an image. You say, “remove the background and save it as a new file” (or whatever) and it happens. Or maybe it happens with an interim verification check in which you review the change the AI directed some app to make and then save the file. Whatever. You get the idea.

This utility reaches an almost exponential explosion of power when all the apps we use can be programmatically controlled this way. But it also gets to a strange point where that app, Paint in this case, is no longer necessary to most as a familiar graphical experience that we directly manipulate. In the short term, Paint and other apps continue forward, of course, and us old-timers will use them as we always have. But increasingly, these apps are no longer apps, and they will start to die off on by one, replaced by what are essentially families or groups of services that each represent individual features. In become programmatic, apps will evolve. They will evolve into something that we, you and I, no longer recognize as apps.

This is happening everywhere.

Last week, OpenAI announced third party app support in ChatGPT, causing some to refer to this AI platform as an operating system. On the face of things, that’s ridiculous, as there are platforms like Teams and Slack that support apps, and they’re not operating systems. But OpenAI’s goal here is the same as that of any middleware maker throughout the ages, like Java or web browsers that deemphasized whatever OS one was using. That is, if ChatGPT reaches its full potential as an interaction point, its users will use it for more and more. And many of its capabilities will come from apps. Apps that will, in time, evolve to be things that are not apps. A series of backend services that platforms like ChatGPT—and Windows—can use programmatically.

So, no, ChatGPT is not an OS and won’t ever become one. But you may use ChatGPT or whatever AI as the central interaction point for accessing personal computing services, some of which were always services and some of which were previously app features. And so it will be a sort of OS, I guess. A front end, similar to how the Windows GUI was once a front-end to the real OS, MS-DOS. And then we’ll see what happens, I guess. There’s no reason to get too far ahead of ourselves, though AI does move quickly.

There’s a lot going on here and this will no doubt trigger future discussions. But for now, I will just leave it at this. As we experience our world changing on an almost daily basis and we freak out every time Microsoft adds some new Copilot-based feature to Notepad or Paint or whatever else, we need to reconcile what it is we’re really seeing. We are bystanders to history in a sense, but what we’re seeing, what’s happening, is perhaps more profound than is immediately obvious. Among all the changes that AI is bringing is this one essential change that many will find distressing. Apps as we know them are on the way out. Apps will eventually die. AI means the end of apps.

This has implications for web browsers, which are multi-faceted apps as it is, but it’s difficult to imagine a world in which they exist in the same form as they are today. Perhaps there will be online “reader” apps and then back-end web services. And “everything” apps like Notion that today seem so versatile will need to evolve into AI systems that orchestrate all those tasks. ChatGPT, Copilot, and other AI apps will likewise evolve into the new “everything” app. Except, of course, that apps will be gone. These things will be experiences, interaction points, or UIs. Whatever you want to call them, they are thin veneers over the services we’re really interacting with, if implicitly.

This is a huge threat to walled gardens like Apple’s App Store and Google’s Play Store because it moves the needle on how we interact with what are today apps. This is similar to how Java and Netscape threatened Microsoft and Windows, as noted. And it is inevitable, which explains why all the major platform makers today are attacking AI so hard.

And one wonders how this changes things in the hardware space, of course. How we will simply walk into rooms and talk to trigger actions. How we will work and consume content and get answers to questions. It’s all changing. But I have to wrap this up somehow. So we will come back to all this. Many times, I think.

We do live in interesting times.