Last Time Was Different
Every technological revolution follows the same pattern. Software broke it. AI restores it.
In This Time Isn't Different, I argued that the technology is genuinely new but the humans and institutions it flows through are the same as they've always been. The pace of transformation is gated by institutional friction, not capability, and that's a feature, not a bug.
This essay is about something else. Not the pace of change, but where the value lands when the change plays out. And the answer, across every major technological revolution, is remarkably consistent — with one glaring exception.
Here's the pattern.
Electricity. GE and Westinghouse made fortunes building the generators and the grid. The companies that used electricity to transform manufacturing, retail, and logistics — they made fortunes too. But the companies in the middle — the ones who built custom electrical installations, wiring systems, switchboard configurations — they had a moment and then got commoditized into oblivion. The value accrued to the people who made the thing and the people who used the thing. Not the people who packaged the thing.
Railroads. The foundational providers (steel companies, locomotive manufacturers) and the end users (companies that used rail to create national distribution networks) captured durable value. The middle layer — the railroad construction companies, the routing specialists — had a boom and then a bust.
Automobiles. Ford and GM made fortunes. The businesses that used cars and trucks to transform logistics, sales, and service delivery made fortunes. The dealership model captured some middle-layer value through regulatory lock-in, but the actual car-customization and outfitting businesses? Commoditized.
The pattern is consistent enough to be a law: in a technological revolution, durable value accrues to foundational providers and to end users who apply the technology to do non-technology things. The middle layer — the people who package the technology for others — captures value only during the window when the technology is too hard for end users to apply directly.
The window opens when the technology arrives. It closes when the technology becomes accessible. Every middle-layer fortune is a bet on that window staying open.
Software broke this pattern.
Microsoft, Oracle, SAP, Salesforce, Adobe — the middle layer didn't just have a moment. It had decades. Enormous, durable, compounding value captured by companies whose entire business was packaging what computers could do for people who couldn't make computers do it themselves.
Microsoft didn't make computers and didn't use computers to do something else. They made the layer between the hardware and the user. Oracle did the same for databases. SAP for manufacturing workflows. Salesforce for sales tracking. The middleware companies. And they didn't just survive — they became some of the most valuable companies in history.
This looked like a new pattern. It looked like technology had matured into a permanent industry where the packaging layer was durably valuable. Career ladders were built on this assumption. Venture capital was structured around it. An entire economic ecosystem — the software industry — formed around the premise that intermediating between "a computer exists" and "a computer does this specific thing for you" was a real, lasting business.
But it wasn't a new pattern. It was the old pattern with an unusually long window.
The window stayed open because software was genuinely hard to build.
That's it. That's the whole explanation. The middle layer captured value for thirty years because the skill required to cross the gap between "computer exists" and "computer does specific thing" was scarce, specialized, and expensive. Unlike electrical wiring or railroad construction, the knowledge didn't diffuse quickly. Programming remained a priesthood. The gap stayed expensive to cross.
And software had a structural amplifier that no previous middle layer enjoyed: zero marginal cost. An electrical contractor had to wire each factory individually. Oracle could sell the same database to a million companies simultaneously. The middle layer's revenue scaled infinitely while the skill moat held. That's why software companies got so big. Not because the middle layer was more durable this time, but because software economics amplified the returns during the window.
The window lasted roughly twenty-five to thirty years — from the SaaS era starting around 2000 to the AI era starting around now. Maybe forty if you count the early enterprise software of the 1980s. That's actually not unusual for a middle-layer window. The electrical contracting boom lasted about the same. It just didn't feel like a window because we were inside it, and the zero-marginal-cost amplifier made the businesses so enormous that they looked permanent.
They weren't permanent. They were the same transitional phenomenon every revolution produces, amplified by software economics into something that looked like an industry but was always a window.
Step back and think about what every software business actually does.
Every one of them — from Salesforce at $300 billion to the micro-SaaS app charging process servers $200 a month — lives in the same gap. The gap between "a computer exists" and "a computer does this specific thing for you." The product is always the same: take a general-purpose computer and make it do a specific thing. Database plus UI plus business logic. CRUD. That's it.
The size of the business correlates with the value of the specific thing. Salesforce is big because tracking sales relationships at enterprise scale is valuable. The process server app is small because tracking legal deliveries is less valuable. But the existence of every one of these businesses depends on the same thing: the gap being expensive to cross.
And sometimes it helps to notice how absurd the gap made things. Salesforce is a three-hundred-billion-dollar company that keeps track of who your salespeople called. That's a database with a UI. The function it performs — "remember who you talked to and what they said" — is the kind of thing that should be, and eventually will be, an inherent capability of a computer. Like a calculator. Like a calendar. Like "type a thing and print it."
At the micro level, the absurdity is sharper. A hundred-thousand-dollar-a-month business selling software to help process servers track their deliveries. The entire function is: here's a list of things to deliver, mark them done, generate a proof of service. That's a list and a form. In what world should that sustain a standalone software business? Only in a world where making a computer do that specific thing requires a specialized intermediary. Only during the window.
Consumers always knew the gap was artificial.
Apple proved it. The only consumer hardware brand that endured understood from the beginning that computer hardware and computer software never really made sense as separate products. You can't buy macOS. In the mind of most consumers, "macOS" only exists as a concept when something goes wrong — an update interrupts them, something crashes, they see a version number they didn't ask about. The rest of the time, it's invisible. It's just how the computer works.
Apple tried to close the entire gap with bundled apps. Pages, Numbers, Keynote, GarageBand, iMovie — the message was: of course your computer lets you type things and print them, of course it lets you edit photos, of course it has a calendar. What kind of computer wouldn't?
The PC ecosystem took the opposite approach. It created the gap by separating hardware from OS from applications — three separate products, three separate revenue streams — and then built an industry in the space between them. You buy a Dell. Then you buy Windows from Microsoft. Then you buy Office from Microsoft again. Then you buy applications from Adobe, Intuit, whoever. Each layer intermediating between you and the thing you actually wanted to do.
Consumers hated this. They always hated it. They tolerated it because they had no choice, but the fragmentation never felt right. When the next wave hit — smartphones — the gap vanished. Apple sold you the device. It did stuff. No separate OS purchase, no separate application ecosystem to navigate. And Google, the other winner of mobile, didn't even sell you a device. They put up the streetlights and maintained them for free so they could sell you the electricity.
The companies that won mobile were the ones that understood computers should just do things. The companies that had built businesses in the gap — Dell, HP, the PC ecosystem — found they had nothing to sell when the gap closed.
AI closes the rest of the gap.
Apple got most of the way there. Your Mac comes with the tools to type, print, edit photos, make music, manage your calendar. But Apple couldn't close the last mile. They couldn't make your Mac do "track my process serving deliveries" or "manage my church's 80 members and their donations" or "tell me which courts I've filed with and when the deadlines are." The specific thing you specifically need still required someone to build custom software.
That's the gap where SaaS lived. The gap Apple couldn't close. Not because Apple wasn't good enough, but because closing it required something that didn't exist yet: a way to go from "I need the computer to do this specific thing" to "the computer does this specific thing" without a specialized intermediary writing code.
AI is that way. When someone can describe what they need and get a working tool — not from the App Store, not from a SaaS vendor, but from the computer itself doing what computers were always supposed to do — the last mile closes. And every business that lived in that last mile loses its reason to exist.
Not because the functions aren't needed. Process servers still need to track deliveries. Churches still need to manage members. Salespeople still need to remember who they called. The functions are as valuable as ever. But the intermediary — the software company that existed because making a computer do the specific thing was hard — has nothing left to intermediate.
So where does value land?
The same place it always lands after a technological revolution.
Foundational providers. The companies building the substrate: Anthropic and OpenAI building the models, NVIDIA building the chips, AWS and Azure running the infrastructure. This is GE and Westinghouse. They make the thing that makes everything else possible.
End users. The people who use AI to do non-AI things. Not "we sell an AI tool" — that's middle layer. "We use AI to serve niches, build products, run businesses in domains that aren't themselves technology." The farmer who used the tractor. The retailer who used the railroad. The manufacturer who used electricity. The value was never in the technology. It was in what you did with it.
The people building AI wrappers, AI-powered SaaS, "AI for X" startups — they're the electrical contractors of this revolution. They'll have a moment. The moment is now. Enjoy it, but don't mistake it for an industry. The margins compress toward zero because the underlying capability is commoditizing faster than anyone can build moats around it. Every "AI for X" startup is a bet that the gap between "AI exists" and "anyone can use AI to do X" stays open long enough to build a business. History says the gap closes. It always closes.
What remains rare — and what every revolution rewards — is knowing what to build and for whom. Taste. Domain knowledge. The ability to see an underserved niche and serve it. The farmer didn't win by selling tractors. He won by farming better. The value wasn't in the technology or in packaging the technology. It was in applying the technology to something real.
The conventional wisdom says "this time is different" — that AI changes everything, that the old rules don't apply. In the companion essay, I argued that the humans and institutions haven't changed, so the pace of transformation follows the old pattern.
But there's a deeper sense in which this time isn't different. The value distribution follows the old pattern too. Foundational providers and end users. Same as electricity. Same as railroads. Same as automobiles. Same as every revolution in which a technology matures from scarce to abundant.
Last time was different. The software industry was the anomaly — a thirty-year window in which the middle layer captured unprecedented value, amplified by zero-marginal-cost economics, sustained by the scarcity of a single skill. That window is closing. Not because the technology is less valuable. Because the skill that kept the window open — the ability to make a computer do a specific thing — is no longer scarce.
Consumers always knew computers should just do things. Apple always knew it. The software industry existed in the gap between that knowledge and reality. The gap was real, and the businesses built in it were real, and the careers built on them were real.
They were also temporary. Same as every other middle-layer fortune in every other technological revolution. We just thought this one was permanent because we were standing in it.