♛
21.07.2026
The Utility Was Never the Story
Read Time: 8 Minutes
Hello
In July 2026, a founder I follow posted that AI can ship 15 items to anyone in 4 weeks, so shipping proves nothing anymore. The real flex is knowing which one of the 15 matters, and having the discipline to ignore the other 14.
I replied with one line. More people are about to learn what it feels like to be a CEO. Someone who can direct capital anywhere, and who has to show restraint and subtract instead. Steve Jobs mode.
The reply kept working on me after I posted it.
Start in 1890.
Electricity
In 1890, electricity was a competitive weapon. If your factory had it and the one across town didn’t, you won. Electricity was the strategy.
By 1930, it was a line item. By 1960, nobody thought about it unless it went out. The technology kept getting more important until it disappeared into the walls. That’s what full adoption looks like. The thing becomes invisible, and being good at it stops mattering, because everyone has it.
Nicholas Carr made the same argument about corporate computing in 2003, in “IT Doesn’t Matter”. People read it as an attack on IT. The actual argument was colder. IT had become infrastructure, available to everyone at similar cost, and something available to everyone at similar cost can’t be the source of anyone’s advantage. Executives hated the essay. It aged perfectly. Nobody in 2026 lists “we use cloud computing” as a differentiator.
Cloud ran the whole arc inside one career. In 2006 it was exotic. Companies debated whether to trust it. Early adopters wrote case studies about their courage. Ten years later it was the default, and by the end of the 2010s running your own servers was the weird choice you had to justify in the board meeting. Electricity took two generations to go invisible. Cloud did it in 15 years.
Now follow the money, because the money did something strange. You’d think the companies that own a transformative technology capture the value of the transformation. Mostly, they don’t.
The railroads transformed America and then went bankrupt in waves. Overbuilt, competed to the bone, hammered by regulation. The value showed up in Sears catalogues, in Chicago meatpacking, in every business that used cheap freight to build something new. The people who moved the goods captured less than the people who figured out what goods to move.
Electric utilities became some of the most regulated, lowest-margin businesses in the economy. The fortunes went to the companies that built things that ran on electricity, and to the brands that organized consumer trust around them.
Cloud looks like the exception. Amazon and Microsoft built massively profitable businesses at the utility layer. Scale and switching costs let them keep margins the railroads never dreamed of. So the strong version of the pattern, that the utility layer always gets squeezed to nothing, is dead. Cloud killed it.
But let’s take a look at what survived. Using the utility gave you nothing. Every company on earth runs on the same few clouds, so no company wins because of it. The durable positions all formed above the layer, in software and marketplaces that hold their ground for reasons that have nothing to do with the servers underneath. Owning the utility can be a great business, and there are maybe 3 seats. Using the utility is never a position. The position lives in what you build on top.
The Factory
Electricity was commercially available by the 1880s. The productivity gains didn’t show up until the 1920s. 40 years of lag. The economist Paul David dug into why, in “The Dynamo and the Computer”, and the answer is almost embarrassing.
Factories were built vertically, around a central steam engine, with power running through shafts and belts. When electricity arrived, factory owners ripped out the steam engine and put a big electric motor in the same spot. Same building, same layout, same workflow, new power source.
They got almost nothing for it.
The gains arrived when a new generation rebuilt the factory itself. Single-story buildings. Machines arranged by workflow instead of by distance from the power shaft. A small motor on every machine. The gains came from the reorganization, and the reorganization took a generation, because it meant scrapping things that still worked.
The winners threw away functioning assets. The losers kept everything that worked and bolted the new thing on.
The Watch
In 1969, Seiko shipped the first quartz watch. Within a decade, quartz made accurate timekeeping essentially free. The Swiss watch industry, which had spent centuries perfecting mechanical accuracy, collapsed. Employment fell from about 90,000 to under 30,000. The quartz crisis nearly ended Swiss watchmaking.
Then the mechanical watch, the technically inferior, wildly expensive, obsolete product, became the luxury. The cost turned out to be the point. When accuracy became free, accuracy stopped meaning anything, and the meaning moved to the visible, verifiable expense. The hand-finishing. The 5 years of a watchmaker’s attention. The refusal to take the cheap path. A Patek Philippe keeps worse time than a $10 Casio, and it costs what it costs because of everything the Casio skips.
The Soap
When mass production made goods abundant in the late 1800s, everything on the shelf looked the same, and you couldn’t tell what was good. Before the factory era, you trusted the person. Your butcher. Your miller. Mass production broke that. The goods now came from a thousand miles away, made by nobody you’d ever meet.
So the market invented a substitute for the person: the brand. A name on the package that stood behind the product, backed by a company with too much to lose to cheat you on one bar of soap. The soap didn’t need the name to be soap. The buyer needed the name to stop checking every bar.
Look at what the buyer actually paid for. Their real constraint was attention, the hours it would take to verify everything personally. The brand sold those hours back. Every trust intermediary since has sold the same product.
When the internet made information abundant, the same problem showed up one level higher. You couldn’t tell what was true. The answer was search, then curation, then reputation systems. Google is a trust business wearing an engineering costume.
Abundance creates a verification problem. The verification problem creates a new intermediary. Every wave.
Now
AI did one thing. It collapsed the cost of producing digital output to basically zero. Text, code, images, analysis, decks, prototypes, campaigns. The marginal cost of one more unit is a rounding error. A person with a laptop can ship what used to take a team a quarter. 15 things in 4 weeks is real.
And the layer is repricing on schedule. The DeepSeek shock of January 2025 was the tell. A frontier-class model from a lab nobody priced in, at a fraction of the assumed cost, and the market repriced the whole layer in a day. Since then, frontier releases get matched fast, and open alternatives keep closing the gap. The direction has held for 18 months. I wrote then that DeepSeek’s real battle was the interface war, and I’ve written about OpenAI as an intelligence utility, because its pricing makes sense as infrastructure economics and stops making sense as product economics.
It was never the models. The question is what just happened to everyone downstream of them.
For all of human history, making something was proof of something. A finished product proved capability. A polished document proved effort. A shipped feature proved a team could execute. Output was a signal because output was expensive.
That signal just died. When a machine produces fluent output for anyone who asks, fluent output tells you nothing about who’s behind it. The cover letter, the pitch deck, the thought-leadership essay, the working demo. Anyone can make all of it, so none of it distinguishes anyone. The signal went to zero the way a currency dies when you print infinite amounts of it. I wrote about the food-supply version of this: cheap tokens are cheap calories. Abundance in the ingredient shows up as bloat in the output. Inboxes, feeds, and pipelines are filling up with industrial-grade fluent nothing.
Industrial agriculture made ingredients cheap and identical for every restaurant on the street. Did restaurants become identical? No. The ingredient became irrelevant to the competition. Once everyone buys from the same commodity supply, no restaurant wins by having tomatoes. They win on the menu. What the chef put on it, and what the chef left off. A 12-item menu is a statement. A 140-item menu is a confession. Tokens are the new produce. Every company now cooks from the same commodity intelligence, so nobody wins by having it, and every company reveals itself by what it chooses to make with it. When intelligence is free, clarity about what to point it at becomes the scarce resource.
Which is what the founder’s tweet was circling. A CEO’s defining condition is the obligation to allocate finite resources against infinite options, publicly, with consequences. Every yes forecloses ten other yeses. That used to be a rare seat. Capital was scarce, so few people ever felt allocation pressure firsthand. AI just gave a version of that seat to everyone. When you can build anything, “what should I build?” lands on you with full weight for the first time. Millions of people are discovering what allocation pressure feels like, and discovering that it’s miserable. Abundance made choosing harder, and it made ducking the choice impossible to hide.
But the “discipline” framing misses a distinction. Difficulty is not cost. When a solo builder declines to ship app number 14, that refusal is psychologically hard and economically free. Nobody sees it. Nothing was at stake. When a CEO kills a revenue line to keep the company coherent, that refusal is expensive and visible. Payroll. Investors. Press. Markets can price the second kind. They can’t even see the first kind.
An announced discipline is a claim, and claims are exactly what just became free. The signal lives in the price of the no.
So refusal is splitting into two markets. The first is external. Refusals priced by observers. Killed product lines, declined customers, exited markets. The expensive, visible kind that analysts and buyers read straight off the financials. This market clears fast. A costly refusal gets repriced within quarters. The second is internal. Refusals priced by time. The solo operator skipping the shiny thing, privately, over and over, with no audience. Nobody prices that refusal today. Compounding prices it. 5 years out, one person has a coherent body of work, and another has 40 abandoned starts, and the gap is the accumulated interest on all those invisible noes. Same mechanism at both scales.
Different clearinghouse.
Different settlement date.
The research backs the mechanism. The famous “paradox of choice,” the idea that more options reliably make people miserable and freeze them, turned out to be shakier than the TED talks suggested. A 2010 meta-analysis by Scheibehenne and colleagues across 50 studies found a mean effect of roughly zero. Chernev’s follow-up work found the real pattern. Choice overload is real and strong under specific conditions, and the big one is when the chooser lacks strong prior preferences.
People who know exactly what they want are fine in front of infinite shelves. People who don’t are wrecked by them.
Translate that. Infinite options tax the filterless. Abundance is a regressive tax, and the exemption is a sharp, pre-existing sense of what you’re for and what you’re not for. The people and companies that wrote their refusals down in advance walk through the infinite shelf untouched. The rest drown in it.
The market has been running the corporate version of this experiment for decades. Diversified firms trade at a persistent discount to focused ones. Around 13% to 15% in the classic Berger and Ofek study. The literature argues about the exact size, and the direction survives the argument. Markets pay a premium for coherence and charge a fee for sprawl. Investors value what you do. They also value what your structure makes it impossible for you to do.
And the scoreboard for all of this already exists. For a company, it’s the P&L. Your P&L is the only positioning statement your company has ever written. Every line of spend is a sentence. Every absence is a refusal. It’s the one document about your company that can’t be generated, because every entry had to be paid for. In-N-Out’s menu has barely changed in decades, and that discipline is legible to anyone who looks. Ford talked about affordable trucks while its capital went elsewhere, and the market read the spend and ignored the speech.
For an individual, the same scoreboard is the calendar and the body of work. What you kept doing, and kept refusing to do, when doing everything was free.
Meanwhile, the big professional services firms are bolting AI onto a pyramid of juniors billing hours for work a model now does. Same building, same layout, same workflow, new power source. The billable hour is the belt-and-shaft system of knowledge work.
What Held
Run the four stories side by side and 5 things have held in every single wave.
One. The arc holds. Every general-purpose technology goes proprietary, then infrastructure, then invisible. No exception in a century. Nothing about AI so far, scaling laws and agents included, has shown any sign of breaking it. Betting on “this time the technology stays a differentiator” has lost every time anyone has made it.
Two. Using the utility is never a position. Owning it can be a fine business, and cloud proved a few of those seats can be enormous. What no wave has ever produced is a company that won because it used the utility. Railroads went broke while Sears got rich. Power utilities got regulated while appliance makers got fortunes. “We have AI” is worth what “we have electricity” was worth in 1960.
Three. Gains come from reorganization. 40 years between electricity and measured productivity, because the gains required scrapping working factories. Adoption is buying the motor. Reorganization is admitting your building is wrong. The first is easy and worthless. The second is brutal, and it’s where all the value lives.
Four. Whatever becomes free, the costly version becomes the luxury. Quartz freed accuracy, and mechanical watches became the asset. Machine production freed goods, and handmade became the premium.
Five. Abundance creates a trust problem, and the trust problem creates a new business. Mass production created brands. Information abundance created search. Every wave ends with a new intermediary whose entire job is telling people what’s real and what’s worth their time.
Underneath all 5 sits the floor.
Attention, time, and trust never became abundant. Not once. Every technology multiplied output, speed, reach, and information. None of them added an hour to the day, a second stream of attention to the human mind, or a shortcut to earned trust. Herbert Simon called it in 1971: a wealth of information creates a poverty of attention. When everything else multiplies, the thing that can’t multiply collects the money. That’s arithmetic, and it has held for a hundred years.
What Happens Next
Six things follow.
“AI-powered” dies as a differentiator. Every pitch deck leading with AI capability is holding a melting asset. The phrase will sound the way “we have a website” sounds now. True and empty. Companies whose identity is “we do X, with AI” will have to answer what they are once the “with AI” clause applies to everyone. A company with no answer will watch its position evaporate on schedule, and the marketing budget won’t slow the melt.
Refusals become the only readable text. When every company’s words, decks, and demos are machine-fluent, the only information that separates anyone is the record of costly decisions. What got funded. What got killed. Which customers got turned away. Analysts, buyers, and hiring managers will get better at reading spend and ignoring speech, because speech is free and spend is priced. The P&L becomes the resume. The calendar becomes the cover letter.
The factory lag picks the winners, invisibly, right now. The AI winners of 2035 are being decided in 2026, by who reorganizes. The firm rebuilding its economics around the model sits next to the firm bolting a model onto its billable hours, and the two look identical in the present tense. Same tools. Same demos. The difference shows up when the lag closes, and by then it can’t be copied, because copying would mean scrapping everything that still works. The sorting will already have happened by the time anyone can see it.
Human judgment becomes the mechanical watch. AI makes fluent thinking free, so costly human attention becomes the luxury good. The median deliverable is already lost to the machine, the way the median watch went quartz. What becomes the Patek is the named human who spent finite, unfakeable hours on your specific problem and signed it. Scarce by construction. Expensive because it refuses to scale. Valuable precisely because of that refusal. Advisory work with a name on the door, small by design, is accidentally built for this future in a way no headcount business can be. The pyramid firms are the Swiss volume manufacturers of 1975.
The next great business sells the hours back. AI’s abundance problem is verification. Provenance, authorship, and whether the claim matches the spend. Somebody builds the layer that tells buyers what’s real. Which company’s stated position matches its capital allocation. Which output had a human behind it. Brands solved this for goods. Search solved it for information. The AI-era equivalent doesn’t fully exist yet, and whoever takes the seat compounds for decades, because trust is the one asset abundance can’t inflate. My own instruments point straight at this seat, so I’m talking my book. I’d make the call anyway.
The middle disappears. Free output plus priced refusal produces a barbell. On one end, massive volume at machine cost. Infinite adequate everything, approaching free. On the other end, scarce, costly, verifiable human commitment at luxury prices. In between sits the competent, undifferentiated professional, firm, and product. Good enough to charge for yesterday. Indistinguishable from the free tier tomorrow. Every industry that digitizes its output gets this shape. Which end of the barbell are you building toward? The middle is a position that stops being available.
What Breaks This
Four things.
The arc breaks. If one lab achieves a runaway capability lead that compounds, a genuine winner-take-all model that never gets matched, then AI has no precedent and the value stays locked in the utility layer for the first time in history. Watch the gap between frontier models and open alternatives. The thesis needs it to keep closing. It has kept closing for 18 months. If it starts widening durably, I’m wrong about the most important premise.
Agents replicate the fixed factor. The floor under everything is that attention and judgment can’t be multiplied. If AI agents become genuinely trusted delegates, the kind you hand a decision to and never review, then attention effectively scales, the floor cracks, and the rent stops flowing to human refusal. I’m skeptical, because delegation without verification recreates the trust problem one level up. You trade checking output for vetting deciders. But I hold this one loosely. It’s the falsifier with the shortest fuse.
Markets start rewarding announced restraint. The whole spine here says claims are free and only costly action signals. If saying the disciplined thing gets priced like doing it, the signalling logic breaks. Every bubble is such a period. They end. Timing is something this framework does not provide.
Costly is necessary and not sufficient. Refusal proves commitment. It doesn’t prove the commitment was smart. You can pay dearly, visibly, and consistently for a position the market simply doesn’t want. The graveyard is full of expensive coherence. This tells you what makes a position believable. It has never told anyone what makes a position correct. Anyone who sells you the second thing wrapped in the first is selling glitter with better lighting.
Finally
Making things was expensive for all of history, so making things meant something. That just ended. The record says what happens next. The technology becomes plumbing. The value moves upstairs. The reorganizers quietly beat the adopters. The costly version of whatever went free becomes the luxury. And somebody builds the trust layer, then collects rent on it for a generation.
Through all of it, 3 things stay scarce. Your hours. Your attention. Whether people believe you. Everything you build should sit on at least one of those, because they’re the only ground that doesn’t inflate.
If you run a company, pull up your last 12 months of spend and read it the way a stranger would. That document is your position. The deck and the website are commentary on it. If the spend reads like a company that couldn’t decide, a rewrite won’t fix it. A refusal will. Kill something that works and doesn’t belong, and let the market watch you do it.
If you’re an individual with these tools, stop polishing output. The machine already won that contest. Write down what you won’t build. Make the list cost you something real. Then hold it long enough for the compounding to show.
In a world where everyone can say anything, the paid-for no is the last sentence anyone believes. The utility was never the story. It never is. The story is what you refuse to do with it.
Access 30+1 Positioning Reports
If any of this was useful
There are three ways to go deeper, depending on where you are.
1. Read and follow.
Everything I publish is free. The Gravity Reports, the Digest, the positioning frameworks — all of it is on the blog and in your feed and inbox.
LinkedIn or X
2. Use the systems.
I’ve built a set of diagnostics for founders and brand leaders who want to do this work on their own company. Start with the free Analyzer — it shows you the gap between what you think you sell and what customers actually buy, and tells you whether AI recommends you when buyers are searching in their language. If you want to go further, the Clarity Kit and Monopoly are built for the same diagnostic, at a deeper level.
Start with the free Analyzer
CEO Clarity Starter Kit
Monopoly
3. Work together.
I take on a small number of advisory clients each year. These are founders and leadership teams who want an outside read on their position — the same method as the Gravity Reports, applied to their own company. If that’s relevant, the best place to start is a conversation. Better call Paul.
DIGEST
