Nobody knows what AI is even for. That's the point.
Everyone agreed what Amazon was for, too.
I’m old. Old enough to have been through multiple tech boom and bust cycles (I started in the dotcom one as a front-end developer). And this time around I’m watching the largest pile of private capital in history, and almost all of the marketing attached to it, pointed at one thing: removing the need for software developers. At the same time I’m seeing the complexity of managing all of this increase (for now??), for example these net new technologies:
OpenRouter for routing your AI requests to the best / cheapest model, and a crazy leaderboard that I’m sure you can have fun placing some weird Kalshi bets on
E2B, a tool that provides enterprises a place to run their agents so they don’t break everything (the OpenAI rogue hacker AIs notwithstanding)
MCPs, the model context protocol that allows AI to connect to existing systems in a way that works for them
Amazon.com makes no money
Which reminds me of Amazon back in the day. When Amazon IPO’ed in 1997, they spent years promising they would be profitable, and weren’t for a long time. I remember thinking that I’d never seen a public company say that and better yet, get away with it. These days they’d likely have just stayed private. But they started by selling books, and then built a massive distribution network and, since loading a truck with other stuff that’s not just books wasn’t super hard, voila, they became the everything store. Around the same time, companies like pets.com, the mascot of dotcom failures of the time, and other retailers like Webvan were tanking horribly.
But then, post-bust in the early 2000s, something else was brewing. Amazon was decentralizing their infrastructure and needed simple programmatic access to it for their own teams so they started slowly building what became AWS. It didn’t hurt that they had learned how to scale massively and had solved for burst problems like Black Friday, leaving them with spare compute the rest of the time. So, they pushed it out to the public as Amazon Web Services, and once launched, anyone could spin up an infinitely scalable internet company without needing capital for the machines, maintenance, data centers, and all the other complexity. They then turned their focus to enterprises and said “we can host your infrastructure better than you can, so let us help you there too”, and it worked.
That leaves us the question of how Amazon actually makes money. In 2025 that was $80 billion in operating income, and more than half of it, about $45.6 billion, came from AWS, from renting out the plumbing they first built for themselves. Most of what’s left is advertising. Amazon doesn’t break out the ad profit, but it books $68 billion in ad revenue at margins some analysts put at around 40%, which is roughly $27 billion. Add that to AWS and you’re at something like nine-tenths of the entire company’s profit. Which leaves the everything store, the trucks and the boxes and the warehouses, as the sliver at the bottom, clearing a margin somewhere around one percent. The bookstore turned out to be a cloud provider with an ad network stapled to the side. Who could have predicted that?
The chatbot is the front door
So what does this mean right now? Stating the obvious, the chat window is not the business. A heavily subsidized $20 per month does not scale to the valuations being thrown around. But that doesn’t mean chats aren’t doing heavy lifting for the AI companies themselves, much like Amazon’s storefront did. It’s the front door and the habit-building machine that’s being created, your “homepage,” so to speak, back when that was a thing worth fighting over. And, even as Amazon has turned their storefront into an ad machine to monetize that surface more profitably, we are still paying something to access these companies’ front doors.
Claude Code is another version of the same game: acquisition and retention of their users. It started as a side project in Boris Cherny’s first week at Anthropic, an AppleScript toy he wired up so Claude could read his music library, and by the time they launched it publicly, over 80% of the company’s own engineers were using it every day. This accrues far more value than a simple subscription alone.
So that brings us to the billion (trillion?) dollar question: what horribly expensive internal problem is so big that someone ends up renting out their solution? If creating AWS was a chore before it was a business, someone is choring (I created a gerund!) as we speak to fix their operational mess (routing, token minning vs maxxing, keeping murderous agents at bay), and it will actually be the product that gets stapled to the front door.
I don’t think anyone knows what the chore is (despite what you may see on X), and that lines up with what I’m hearing from both entrepreneurs and enterprises alike. But it won’t be categorized as an AI thing per se, just as AWS has nothing to do with selling books.
Somewhere in the mess
So “replacing all developers” was the shiny thing to look at but, if I’m right, like all shiny things in this space, it likely means we’re looking at the wrong thing, he says while staring at his metaverse NFT. I’m also well aware of the fact that the internet was based on open, free protocols (HTTP, DNS, TCP/IP, SMTP) that were publicly funded, and that AI is privately built and funded right down to the Nvidia metal. Even the open-weight models that look like a gift can be un-gifted (see Meta stopping right before Behemoth in April of 2025). So the analogy isn’t perfect, making it even harder to get it right. But, like all of history, it rhymes.
Somewhere in this mess is a business model that is gestating, despite all the hype and what is being claimed with certainty. It may emerge from the ashes of startups being built right now. What I can say with certainty is that it’s not what we’re all staring at, as I write this on my Aeron chair, acquired for free from the last pile of ashes with the wrong business model.
AI4
Speaking of nobody knowing the chore: next week I’ll be at AI4 in Las Vegas (August 4-6), in a room full business leaders trying to untangle AI. So here’s a real ask: What do you actually want to know about how big companies are thinking about AI? What would you put to them if you had the badge and the room? Hit reply and tell me. I’ll carry your questions in and report back what I actually find, not what I hear in the keynotes. Get them to me before Aug 3 and I’ll go hunting.
Where You’ll Find Me
AI4, Las Vegas, August 4-6. Come say hi if you’re there. I’m always up for a hallway conversation about which business model is currently on fire.
SaaS North, Ottawa, November 3-5. I’ll be hosting a shin-dig at this event, subscribers get dibs. Stay tuned!