Intellification

Contents

1. An afternoon in August
2. A word for what just happened
3. Everyone gets the first decade wrong
4. Group drive: the mistake we are making right now
5. Unit drive: what actually paid
6. The application is a line shaft
7. Reading the ICE-to-EV map
8. There is no software company
9. Five rules for owning something nobody sells
10. How long this takes

A man who owns three service trucks sits down at his laptop after lunch and types one sentence: build my business a phone receptionist. Then he adds the part that matters. Don’t ask me for permission. Just do it. He closes the laptop and drives to a job site.

He did not buy software. That is the whole point, and almost nobody has said it out loud yet.

There is no vendor. No dashboard he logs into. No seat license, no onboarding call, no roadmap, no account manager. If he wants the receptionist to stop offering Thursdays, he cannot file a feature request, because there is no one to file it with. He goes back to the machine and says so.

Section 01A word for what just happened

We call the last great transition electrification, and the word is a small lie. It describes a swap — steam out, electricity in — and the swap is the least interesting thing that happened. What actually happened is that once power could be delivered anywhere, cheaply, in any quantity, the building changed shape. The machine that made the power stopped being the thing you designed around.

I want a word for the same event happening to software, so: intellification. The transition from buying bundled capability to dispatching intelligence. From a thing you log into, to a thing you tell.

The parallel that makes this legible to almost anyone is the one sitting in their driveway. Internal combustion is to software what the electric vehicle is to AI. Both old regimes are mature, enormously capable, deeply embedded, serviced by a whole professional class, and structurally organized around a scarcity that no longer exists.

Figure 01

What an afternoon looks like when you don’t ask for permission

1. The instruction

   One sentence, plus explicit authority to act without checking back. The authority is the load-bearing part.

2. A phone number is leased

   A US local number on Twilio runs $1.15 a month; inbound calls to it are $0.0085 a minute. The front door of the business now costs about a coffee a year.

3. The wiring

   Webhooks, API keys, a speech pipeline, a place to run it. Every one of these is a word the owner does not know and will never need to.

4. The knowledge

   Hours, service area, pricing floor, what counts as an emergency, what to never promise. This is the only part that is actually his.

5. The write-back

   An appointment lands in the CRM. The loop closes, and the thing stops being a demo.

6. The part nobody built

   A place to see what was signed up for, on whose card, under whose name. This step is missing, and its absence is the real subject of this essay.

Section 02Everyone gets the first decade wrong

In 1990 the Stanford economist Paul David wrote a short paper that has been quietly correct for thirty-six years. Economists at the time were baffled: computers were everywhere, productivity statistics were flat. Robert Solow had put it best — we see the computers everywhere but in the productivity statistics.

David’s answer was to go back to the dynamo. In 1899, seventeen years after Edison lit the first central generating stations in London and New York, electric motors accounted for less than five percent of American factory mechanical drive. Electric lighting was in three percent of homes. Not because electricity was bad. Because factories already worked.

In 1900, contemporary observers well might have remarked that the electric dynamos were to be seen everywhere but in the productivity statistics.

It took forty years from the first central station before even half the mechanical drive in American factories was electrified. Manufacturing productivity did not accelerate until the 1920s. When it did, David found that roughly half of a five-percentage-point jump in total factor productivity growth was statistically accounted for by one variable: the growth in installed secondary electric motor capacity between 1919 and 1929.

Secondary motors. Not motors. That distinction is the entire essay.

Figure 02

Forty years from the first central station to half the factory floor

4.8%

24.7%

53.1%

78.4%

86.1%

• 1899
• 1909
• 1919
• 1929
• 1939

Section 03Group drive: the mistake we are making right now

Here is what a factory did when it first electrified. It bought an electric motor, bolted it where the steam engine had been, and connected it to the same overhead line shaft that had always been there. The same belts came down from the ceiling to the same machines in the same multi-story building. This was called group drive, and it was the dominant approach from the mid-1890s to the eve of the 1920s.

It worked. It saved some fuel and gave better speed control. It also, as David points out with real economist’s relish, made the productivity numbers worse — because the old belts, shafts and engines stayed in place as available capacity, which raised the capital-to-output ratio while delivering gains that the statistics did not capture.

A new power source, wired into the old transmission system, produces a modest improvement and a confusing balance sheet. Hold that thought and look at what most companies bought this year.

The chat box in the corner of the CRM. The “AI-powered” tab in the accounting suite. The summarize button on the ticketing system. Every one of these is a motor bolted to a line shaft. Real intelligence, delivered through a transmission apparatus — the application, the seat, the login, the vendor — that exists for reasons which no longer apply.

Figure 03

Two ways to wire the same power

Group drive · 1895–1920 · and most software today

One source, distributed through fixed apparatus

The motor replaces the engine. Everything downstream stays: the shafts, the belts, the bracing, the multi-story building erected to keep the shafts short. To change one machine you shut down the section.

• Line shafts
• The application
• Belt housings
• Per-seat licenses
• Overhead bracing
• The vendor roadmap

↓  the apparatus goes away  ↓

Unit drive · 1920s · and what is starting now

Power at the point of use, in whatever quantity that point needs

Each machine gets its own motor. The building can now be single-story, lighter, and rearranged on a weekend. Nothing has to be shut down to change one thing.

• Single-story plants
• One job, one build
• Flexible layouts
• No seats
• No full shutdowns
• Instructions, not tickets

Section 04Unit drive: what actually paid

The thing that finally moved the productivity needle was not the motor. It was that once each machine had its own small motor, the factory did not need line shafts at all — and once you do not need line shafts, everything you built to support them becomes unnecessary.

David lists the consequences and they are startling in how little they have to do with electricity. Buildings got lighter, because nothing had to carry the weight of overhead transmission. They went from multi-story to single-story, because the old height had been dictated by the need to keep line shafts short. Layouts could be optimized around how materials actually moved, and rearranged when the product changed. Maintenance stopped requiring a full shutdown, because the power system was modular.

The dynamo’s payoff was not in the motor. It was in the floor plan that only became drawable once you had one.

And the reason it took twenty-five years is that somebody had to learn how to draw that floor plan. David is precise about this: the knowledge lived in factory architects and electrical engineers, and the industry that employed them was fragmented and had high turnover, so the learning could not accumulate. The bottleneck was never the technology. It was the cadre.

Section 05The application is a line shaft

Software’s line shaft is the application — a bundle of capability arranged around a shared interface, sold as a seat, priced monthly, and maintained by a company.

Ask why that bundle exists at all and the answer is entirely about cost. Building capability was expensive, so the cost had to be spread across thousands of customers. Spreading it required making one artifact that many businesses could use, which required generality, which required configuration, which required a user interface, which required documentation, support, onboarding and a roadmap for all the things it does not do yet. Every layer of that stack is transmission apparatus. None of it is the power.

Gartner expects the world to spend about $1.47 trillion on software this year, growing more than fifteen percent — the largest single-year expansion of software spend on record. That is a group-drive number. It is what it looks like when a new power source gets wired into an old apparatus at scale: spending goes up, because you are now paying for both.

The receptionist answering that man’s phone has one customer. It has no configuration screen because it has nothing to configure — it was built for him. It has no roadmap because there is no queue of other customers to prioritize against. It has no pricing page. It is unit drive.

Section 06Reading the ICE-to-EV map

If the factory analogy is too far back, use the one in the driveway. The transition from internal combustion to electric is the same structural event, and it is far enough along to argue from evidence instead of hope. One in four new cars sold in the world last year was electric — more than twenty million of them. In China it was nearly fifty-five percent. In the United States, just under ten.

Figure 04

Electric share of new car sales, 2025

• China~55%
• Europe28%
• World25%
• Southeast Asia~20%
• United States<10%

• Purple — national and regional markets
• Cyan — global average

Notice the shape of the American number. Ten percent, in the country that invented most of this, while China is at fifty-five. Adoption is not a function of whether the technology works. It is a function of price, policy, and whether the thing you already own still runs. Exactly what stalled the dynamo for forty years.

Section 07There is no software company

Now the part that should make you slightly uncomfortable, because it makes me slightly uncomfortable.

The receptionist is answering calls from that man’s customers. It is speaking on behalf of his business, quoting his terms, and writing to his records. It was assembled by a machine, out of components he cannot name, on accounts he has never logged into, funded by a card charge he will see next month with a line item that says TWILIO.

He cannot call anyone. If it starts telling people he does emergency service at 2 a.m. when he does not, there is no vendor to escalate to, no status page, no SLA. There is him, and the machine, and whatever he can remember about what he asked for.

This is not a reason to stop. It is a reason to understand what he now owns, because he owns something genuinely new: an operating asset with no manufacturer. Fifty-eight percent of American small businesses were already using generative AI as of the Chamber of Commerce’s last count — up from twenty-three percent two years earlier — and almost none of them have thought about this.

Section 08Five rules for owning something nobody sells

1. Own the accounts, not the artifact

   The phone number, the domain, the CRM, the API keys, the card they bill to — in your name, your email, your control. The build is disposable; it can be regenerated in an afternoon. The accounts cannot. This is the new title deed, and it is the only thing in the whole arrangement that is actually property.

2. Keep the instruction, not the code

   Whatever you told it is the thing you can rebuild from. Nobody needs to read the code, including you. Save the sentences in a plain text file and keep it wherever you keep your operating agreement. That file is the specification for your business.

3. Your card statement is your bill of materials

   You no longer have a vendor list, because you no longer have vendors. What you have is line items. Read them monthly. It is now the only complete inventory of what your company depends on to function.

4. Call your own number

   Every week. Ask it something awkward. Ask it something you would never authorize. You are the QA department now, and that is not a burden — it is four minutes, and it is the single highest-leverage four minutes in your week.

5. Design for a confident mistake, not an outage

   Software fails loudly: it errors, it stops, you find out. Intelligence fails plausibly — it keeps answering in the same calm voice and is simply wrong. Nothing crashes. Nothing alerts. Build your checks for the failure mode you actually have.

Section 09How long this takes

Forty years from Pearl Street to half the factory floor. That is the number people reach for when they want to argue this will all take longer than the enthusiasts say, and they are usually right to reach for it.

But read David’s actual explanation for the lag. It was stranded capital: you could not justify tearing out a serviceable steam plant and a still-solid multi-story mill to chase a benefit you could not yet quantify. The delay was not caused by doubt. It was caused by depreciation schedules.

That constraint is mostly gone. The stranded capital in our man’s case was a two-hundred-dollar-a-month answering service and one afternoon. Nobody has to wait for a building to wear out.

David’s second explanation has not gone anywhere at all. Somebody had to learn to draw the new floor plan, and the knowledge could not accumulate because it sat in a fragmented industry that could not hold onto it. That is precisely where we are. Nobody knows what an intellified small business looks like, because almost none exist yet. There is no cadre. The guy with three trucks is the cadre, and he is learning it by doing it, which is the only way it has ever been learned.

Already true

A working business phone line costs $1.15 a month plus under a cent a minute. One in four new cars sold worldwide last year was electric. Fifty-eight percent of US small businesses report using generative AI. None of this is forecast; all of it is on the record.

What has to happen next

The accounts layer has to become boring. One ordinary place where an owner can see every service an agent signed him up for, on whose card, under whose name, with a switch to turn each one off. That is not a model problem and it is not hard. It is just unglamorous, and nobody has built it.

Where I am probably wrong

I think the software bundle comes apart fast. It might not, for one reason I keep circling: the bundle was never only about build cost. It was also about accountability — someone to sue, someone to certify, someone whose name goes on the compliance form. If liability turns out to be the load-bearing part of the application, then intellification stops cleanly at the size of business where nobody asks that question. My man with three trucks would be the end of the story rather than the beginning of it, and I would have written this essay about a niche.

Either way, the question to stop asking is what software should I buy. That question assumes the answer is a product, and increasingly it isn’t. The question underneath it is: what job do I want done, and whose name is on the account it runs on. The first half is the work. The second half is the ownership. There is no longer very much in between, and what remains in between is being sold to you at fifteen percent annual growth.

He still cannot tell you what Twilio is. That is fine. The men who ran the first unit-drive factories could not have explained a magnetic field either. What they could tell you was that the room was brighter, the floor was clear, and you no longer had to stop the whole mill to move one machine. That was enough to change how every building in America got drawn.

It is still enough.