"What is happening with agents is that the ability to do stuff is becoming abundant; what is scarce is volition." — Ben Thompson, Stratechery, Sept. 28, 2026
Your phone has 689 apps. You use maybe nine. That gap is not a you problem — it's the whole story of the next five years, and Ben Thompson just told it in one sentence: agents aren't an app, they're the thing that makes apps irrelevant.
Here's the fun part: the receipts are already in.
Thompson's argument, from his article Apps, Agents, and Aggregation, is almost embarrassingly simple once you see it. For 20 years the internet's scarce resource was discovery — finding the site, the app, the thing. Whoever solved discovery (Google, Meta) aggregated demand and got rich. That's Aggregation Theory, and it's why Zuck could say "apps aren't the center of the world" back in 2013.
Now flip it. Distribution is free, doing things is easy, and the scarce resource has moved. It's volition — actually wanting to get something done, and handing that want to something that can execute it. When the agent is the thing you talk to, every app becomes an implementation detail. A supplier. A nameless subcontractor that the agent hires and fires without you ever seeing the logo.
In 2013, Thompson counted his home screens: 151 apps. He published the screenshots to argue with Zuckerberg.
Today? 689 apps. And the honest confession buried in the piece: he opens a shrinking handful of them. Messaging, a social app, and ChatGPT. The other 680 are a museum of "things I once needed to do." Nobody wants to use apps for their own sake. Apps were always just the vehicle.
Then he did the demos that make this real. He asked Meta's Muse: "Can you organize all of the recipes I've saved in Instagram?" He walked the dog. By the time he got home — about 5 minutes — he had a custom app: Recipe Box, app number 690.
He's calling it "the infinite app," because the agent can conjure one for any job, on demand, for an audience of exactly one. And here's the infrastructure that makes it possible, which almost nobody noticed: Meta is handing every U.S. user a real virtual machine — 2-core processor, 8GB of RAM, 8GB of storage — so their agent has its own computer. You don't just get an agent. You get a computer for the agent.
If agents are the new gatekeepers, who gets to stand at the gate? Ask Amazon.
On Sept. 20, 2026 — twelve days after Muse launched — Amazon blocked Meta's agent from checking out on Amazon.com, citing unauthorized automated access and credential concerns. Shopify shrugged and welcomed it. Walmart, Best Buy, Wayfair, Sephora, and Gap signed up as partners; Shop Pay and PayPal became payment rails. Meta's business model is a toll booth: the agent is free, and Meta skims a small fee off transactions it completes.
This is the oldest fight in retail — Amazon vs. Walmart — wearing an AI costume. Only now the battlefield is who owns the customer's account, and the customer might never see the checkout page at all.
Once you look for it, you can't unsee the convergence:
Three companies, three logos, one architecture. And Salesforce is reportedly trying to charge a three-digit premium to make its flagship product a plug-in for Claude — a valiant attempt to sell toll tickets to a road that's about to be paved over.
Then there's Meta Enterprise Platform, announced a day after Thompson's article. His verdict was blunt: "who is the customer?" — and Meta has a once-in-a-generation shot at the consumer agent layer, so chasing enterprise is a distraction from the biggest prize in consumer tech. The market's reaction (a shrug, at best) suggests investors agree.
The logic: agents get better the more context and access they have, which makes them sticky — and most people will have exactly one. Meta's distribution is nearly every human on Earth. Microsoft's is nearly every employee. Those two facts, not model quality, are the moat.
This isn't vibes. Morgan Stanley puts agentic commerce at $190–385 billion in U.S. e-commerce by 2030 — 10–20% of the market. The retail agentic-AI segment ran about $46.7 billion in 2025 and is forecast to hit $175 billion by 2030 (~30% CAGR). And 23% of Americans say they bought something via AI in the past month, with AI-referred traffic to U.S. retail sites up a staggering 805% year-over-year on Black Friday 2025.
The user stops browsing. The agent starts buying. The app store becomes a supply closet.
Ben Thompson's provocation is worth quoting twice: apps were never the point. They were always a means to a job to be done. Agents do the job. The companies that own the "want" own the world — and right now Meta, Microsoft, and OpenAI are racing to be the one you talk to. Your 689 apps? They're already résumés sitting in the inbox of a boss who hasn't interviewed them yet.
Sources: Stratechery — "Apps, Agents, and Aggregation" (Sept. 28, 2026) and "One More Note on Agents, Meta Connect, Meta Enterprise Platform" (Sept. 29, 2026); Stratechery — "This Week in Stratechery" (Oct. 2, 2026); OpenAI Dev Day 2026 recap; The Verge / Wired / Platformer on OpenAI dots; GeekWire, Forbes, NBC News, MarketBeat on Amazon blocking Meta's Muse; Morningstar and Hyperframe Research on Meta Enterprise Platform; Morgan Stanley agentic-commerce outlook; Mordor Intelligence and Nevermined agentic-retail market data.
