
Apps, Agents, and Aggregation
Three revolutionary products — you know the line. A wide-screen iPod with touch controls, a revolutionary mobile phone, and a breakthrough Internet communications device. That was how Steve Jobs introduced the iPhone:
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Three revolutionary products — you know the line. A wide-screen iPod with touch controls, a revolutionary mobile phone, and a breakthrough Internet communications device. That was how Steve Jobs introduced the iPhone:
Today, we’re introducing three revolutionary products of this class. The first one is a wide-screen iPod with touch controls. The second is a revolutionary mobile phone. The third is a breakthrough Internet communications device. So, three things: a widescreen iPod with touch controls, a revolutionary mobile phone, and a breakthrough Internet communications device. An iPod. A phone. And an Internet communicator. An iPod. A phone. Are you getting it? These are not three separate devices, this is one device, and we are calling it iPhone. Today, Apple is going to reinvent the phone.
I’ve linked to this snippet before, usually to note how the audience didn’t really understand what an “Internet communications device” was, even though that was the iPhone’s most revolutionary capability.
The phrase I’m thinking of right now, however, is “Are you getting it?” Are you getting that messaging — chatbots now, natural interfaces later — is how we will communicate with AI? Are you getting that not only will we not program computers, we won’t use them — AI will? Are you getting that pre-built UI — write once, run everywhere, for everyone — is dead? These are not three separate predictions: this is reality, right now, in 2026. The future is here, even if it’s not widely distributed. Or is it?
I wrote on February 18, 2014 that Messaging was Mobile’s Killer App; I specify the date because I fortuitously published that Article the day before Facebook acquired WhatsApp. My thesis was that phones may have started out as convenient information devices, but the killer use case was a very human one:
Seven years ago, the computer became pocketable, but the original use cases were about making the passive presentation of information accessible not just at a time convenient to the viewer, but also at any place: the web was now everywhere. Still, it’s only recently that the killer app for this era, when the nodes of communication are smartphones, has become apparent, and it is messaging. While the home telephone enabled real-time communication, and the web passive communication, messaging enables constant communication. Conversations are never ending, and friends come and go at a pace dictated not by physicality, but rather by attention. And, given that we are all humans and crave human interaction and affection, we are more than happy to give massive amounts of attention to messaging, to those who matter most to us, and who are always there in our pockets and purses.
Meta’s WhatsApp acquisition was actually a step back in ambition; a year earlier the company had launched its own interface for Android called Facebook Home; CEO Mark Zuckerberg told Wired at the time:
Home turns your phone into a Facebook device. Even with the lock screen on, a photo stream of your friends’ activities fills the screen. Updates appear on your home screen, too. What’s more, Home makes Facebook the primary means of communication on your device. The company’s messaging software merges with SMS, and you can continue using its “chat heads” to text while inside another app. Zuckerberg believes that the social network plays too big a role in its users lives to be drowned out by a vast sea of apps. “Apps aren’t the center of the world,” he says. “People are.”
I stridently disagreed, writing in Apps, People, and Jobs to Be Done:
Apps versus People. According to Zuckerberg, that’s the dichotomy. And he’s wrong. He forgot about jobs to be done.
Here are my (carefully curated) home screens:
Focus on how many of the icons are about “People”…four in total across three screens. People matter to me, but I use my phone for so much more.
So what if I consider “Jobs to be Done”?…Total tally: 151 apps, 80 jobs to be done, 4 foci on people
Apps aren’t the center of the world…But neither are people. The reason why smartphones rule the world is because they do more jobs for more people in more places than anything in the history of mankind. Facebook Home makes jobs harder to do, in effect demoting them to the folders on my third screen. Who is Facebook to prioritize my jobs?
Those screenshots show 151 apps; today my phone has 689. No, I rarely open most of them, but hey, if I again need the random app I downloaded for whatever reason in the past, it’s there. More generally, those numbers speak to what the iPhone and the App Store made possible: customized interfaces for a whole host of activities that were previously inaccessible on the go.
Still, the core set of apps has remained remarkably consistent: messaging apps, including WhatsApp, social media, primarily X, and now ChatGPT and Claude.
I got a new app yesterday, inspired, coincidentally enough, by a WhatsApp group chat.
I have been all-in on agents for a while. First coding agents, then an overall idea tracker agent I fashioned from a dedicated Claude thread, and most recently an agent specifically tuned to my company’s needs. It was that experience that made me instantly excited about Muse, which actually gave people the infrastructure necessary to have an agent of their own.
The key thing to understand about agents is that they are not just AI: they are an AI that has access to a computer. It was clear very early on in the ChatGPT era that AI would not replace computers, but rather operate them; from 2023’s ChatGPT Gets a Computer, on the occasion of ChatGPT adding plugins, including one from Wolfram|Alpha:
The fact this works so well is itself a testament to what Assistant AI’s are, and are not: they are not computing as we have previously understood it; they are shockingly human in their way of “thinking” and communicating. And frankly, I would have had a hard time solving those three questions as well — that’s what computers are for! And now ChatGPT has a computer of its own.
The idea was not that probabilistic LLMs would somehow become deterministic, but rather that LLMs would be able to leverage deterministic computers to do computer things; that’s exactly what an agent does, augmented by its ability to write things down. It follows, then, that to actually deliver an agent that is usable by people unable or unwilling to set up a computer for their agent, you have to give people not just an agent but also a computer for their agent.
This is why the aspect of the Muse launch I latched onto was not the Muse Spark model that undergirded Muse, but rather the fact that Meta was provisioning every user in the U.S. (and presumably, eventually the world) with a virtual machine with a 2-core processor, 8GB of RAM, and 8GB of storage. That’s a real-deal computer, which is pretty remarkable, and also the only way to make agents work for most people.
Of course the question remains what you actually use this computer for, which is where my new app comes in: after hearing about how a friend used Muse to look at his Instagram favorites, I asked Muse:
Can you organize all of the recipes I’ve saved in Instagram?
After first delivering a PDF (“Hmm, a PDF isn’t very useful. You can’t store them in a more accessible format?”), I got myself a new app:
It took 5 minutes, and the entire conversation took place while I was walking my dog.
The app’s not perfect: right now it’s really a way to categorize Instagram videos; I asked Muse to actually read the captions and watch the videos; it’s working on it, subject to Instagram rate-limiting:
I can be patient; I have 689 other apps to look at in the meantime.
Of course I’m not going to do that; in fact, the number of apps I look at are plummeting. I understand why the idea of giving agents access to your own computer is scary, but every app with a command-line has been usable by agents for a long time; starting with GPT-5.6 Sol, every app with a user interface was usable too, albeit slowly. Astra made it fast. And, well, that’s it: AI can basically use any app or any website that I don’t want to.
And frankly, I’m fine with it. The reason why I had 689 apps was because I had that many services or games that I did, at least at some point, want or need to interact with; all of them required learning how they worked to get done what I wanted to get done. That, after all, was the point: I’m not a professional app user, I’m just someone who wants to get things done. And AI gets it done.
I framed my new Muse-built Recipe Box app as number 690 on my phone. In fact, however, it represents the infinite app: the agent of my choosing can create any app that I want on command, even if that app is only for me. This is the earliest manifestation of another future that has been clear with the rise of LLMs: truly customized UI. From 2024’s The Gen AI Bridge to the Future:
This is where you start to see the bridge: what I am describing is an application of generative AI, specifically to on-demand UI interfaces. It’s also an application that you can imagine being useful on devices that already exist. A watch application, for example, would be much more usable if, instead of trying to navigate by touch like a small iPhone, it could simply show you the exact choices you need to make at a specific moment in time. Again, we get hints of that today through deterministic programming, but the ultimate application will be on-demand via generative AI.
Of course generative AI is also usable on the phone, and that is where I expect most of the exploration around generative UI to happen for now. We certainly see plenty of experimentation and rapid development of generative AI broadly, just as we saw plenty of experimentation and rapid development of the Internet on PCs. That experimentation and development was not just usable on the PC, but it also created the bridge to the smartphone; I think that generative AI is doing the same thing in terms of building a bridge to wearables that are not accessories, but general purpose computers in their own right.
Last week’s Meta Connect was, as it is every year, about new Meta devices, primarily glasses; what was notable about this year, however, was how Muse made everything make sense. Meta’s devices are not AR or VR devices, or even AI glasses: they are Muse delivery mechanisms. Indeed, the general purpose computer I was talking about exists: Meta gave it to every user for Muse to use.
And, as we saw with my recipe box app, generative UI is here. No, it’s not the just-in-time UI that I do still think is coming, but it’s certainly on that path: I had custom UI generated just for me, with no need that the app be shared with anyone else. It’s effectively disposable, because it’s infinite.
This is a replay of what happened with websites. Publications used to be scarce, dependent as they were on printing presses and delivery trucks; then, distribution became free, and the web became abundant. That, by extension, meant the most important problem to be solved was discovery; the companies that solved discovery aggregated demand, giving them power over suppliers, and in the case of Meta and Google in particular, incredible advertising businesses (this is Aggregation Theory).
What remained scarce, however, was actually doing things on the web, or in apps. What is happening with agents is that the ability to do stuff is becoming abundant; what is scarce is volition. The problem to be solved is not discovery, but rather inspiration; the companies who solve inspiration will gain power over every entity that has things that need to be done. Once the user is focused on solving a problem, every app and service required to do so is abstracted away into an implementation detail, mere suppliers facing the fate of publications under Aggregators, scrapping for crumbs from the Agent, the ultimate gatekeeper of not just user demand, but desire.
After teasing its new Copilot “super app” last month, Microsoft is offi
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。