The Generator Was Never the Moat
This essay is about AI -- and why software moats don't disappear just because agents can generate code.
Samuel Insull was exhausted.
For three years, he'd been running Chicago Edison at a pace that would have killed most men. He'd inherited a money-losing utility in a city with 29 competing power franchises, some covering entire districts, others just "a few blocks on the northwest side." The business was chaos. Customers paid by the number of bulbs installed, not by usage. Generators were small and inefficient. And everyone in the industry was terrified.
The terror had a name: alternating current.
Insull had spent a decade as Thomas Edison's personal secretary--his gatekeeper, confidant, and operational right hand. He'd watched Edison fight the War of Currents with religious fervor, publicly electrocuting dogs and horses to prove that AC was the "executioner's current." He'd watched Edison lose. By 1893, Westinghouse's alternating current had won the contract to light the Chicago World's Fair. The technical debate was over.
Edison's loyalists believed this meant the end of defensible electric businesses. If anyone could generate alternating current using standard Westinghouse equipment, what was the moat? The generator--Edison's great invention--had become a commodity. The technology was standardizing. The future looked like a race to the bottom.
So in the winter of 1894, worn out and uncertain, Insull did something unusual: he took a vacation. He went to Brighton, on the English coast, to rest.
What he saw there changed everything.
As evening fell, the seaside town lit up. Every shop, no matter how small, blazed with electric light. In America, businesses turned off their lights the moment they closed--electricity was too expensive to waste on empty storefronts. But in Brighton, the lights burned all night.
Insull found the manager of the local power plant and demanded an explanation. The answer was simple: Brighton had invented a meter that measured actual electricity consumption. Instead of charging by the bulb, they charged by usage--with different rates for peak and off-peak hours. By making nighttime electricity cheap, they'd convinced shopkeepers to leave their lights on. And by smoothing demand across the day, they'd transformed the economics of the entire business.
Insull returned to Chicago with an insight that would make him one of the richest men in America:
The generator was never the moat.
The moat was the wire to every home. The meter on every wall. The billing relationship with every customer. The franchise agreement with every city. The infrastructure so deeply embedded in daily life that ripping it out was unthinkable.
He stopped fighting AC. He adopted it. He bought up competitors. He vertically integrated--owning the coal mines, the railroads, the barges, and the power plants. He lobbied for regulated monopolies, arguing that "competition is an unsound economic regulator" for a business with such enormous fixed costs. By the 1920s, 95% of Chicago homes were wired for electricity. His empire controlled power companies in 32 states. The optimistic view when he'd started was that "as many as 25,000 Chicagoans might ultimately use electricity." He ended up serving millions.
The technology had commoditized. The business had compounded.
On the morning of March 1, 2026--131 years after Insull's Brighton epiphany--Naval Ravikant posted nine words to X:
"Pure software is rapidly becoming un-investable."
Within hours: 3.4 million impressions, 20,000 likes, and a discourse firestorm. Quote tweets from VCs, founders, and AI researchers piled up--some nodding gravely, others firing back.
Naval's argument is the same argument Edison's loyalists made in 1892. If the technology is commoditizing--if anyone can spin up an LLM wrapper, if the marginal cost of intelligence is collapsing, if AI can generate code as easily as Westinghouse could generate alternating current--then what exactly are you investing in?
The anxiety is real. And Naval has a track record of seeing inflection points early.
But Insull's lesson suggests he's looking at the wrong layer. The question isn't whether the generator is becoming a commodity. The question is: who controls the wires?
In electricity, the wires were literal: copper cables running from generating stations to every home, every factory, every streetlight in the city. Insull understood that once those wires were installed, once the meters were mounted, once the billing relationships were established, the customer was his. The generator could be built by anyone. The wire to the customer's kitchen could not.
In software, the wires are network effects.
Figma didn't win the design tool market by having better features than Sketch. It won by building Multi-Cursor Presence--turning design from a single-player tool into a real-time collaboration platform. Every additional teammate made the workspace more valuable. Every shared component library deepened the dependency. Every AI feature that integrated into those shared files strengthened the network rather than weakening it.
The network became the wire.
AI doesn't eliminate these network effects. It amplifies them. Products with the most users generate the most interaction data. Products with the most interaction data train the best models. Products with the best models deliver the best outcomes.
Switching isn't just abandoning features anymore--it's abandoning context, collaboration, and compounding intelligence.
Insull's customers couldn't easily switch power companies because the wires were buried in their walls. Figma's customers can't easily switch design tools because the network is embedded in their workflows.
The generator was never the moat. The wires were.
Insull's Brighton meter didn't just measure electricity. It created a relationship.
Before the meter, electricity was a luxury product with flat pricing--you paid by the bulb, whether you used it or not. After the meter, electricity became a utility with usage-based pricing that aligned the interests of provider and customer. The meter gave customers visibility into their consumption. It allowed Insull to offer off-peak discounts that increased utilization. It transformed a one-time installation into an ongoing commercial relationship.
In software, the meter is customer success.
The strongest SaaS businesses have always been built on net revenue retention--the percentage of recurring revenue that survives plus expansion from existing customers. This metric measures how deeply the product is wired into customer operations, how much value it delivers over time, and how hard it would be to rip out.
| Metric | Value |
|---|---|
| Median NRR ($10-50M ARR SaaS) | 105-110% |
| Top Quartile NRR | 115%+ |
AI makes these relationships stickier, not weaker. AI-powered products that learn customer preferences, automate customer workflows, and predict customer needs create dependencies that static software never could.
Every automated workflow becomes part of the customer's operational fabric. Every LLM-driven insight becomes part of their decision-making process. Every AI-generated asset becomes part of their institutional memory.
The meter on the wall made the power company irreplaceable. The customer success motion makes the software irreplaceable.
Insull's power grid required constant maintenance. Generators broke. Lines fell. Demand fluctuated unpredictably. The system needed continuous human intervention to stay operational--an army of engineers, linemen, and operators working around the clock.
AI-native software is building something Insull could only dream of: a grid that heals itself.
Modern infrastructure can flag anomalies, auto-remediate issues, rewrite inefficient queries, reallocate compute dynamically, and optimize for new usage patterns--without human intervention. The marginal cost of maintaining software is collapsing even as the systems become more sophisticated.
And the cost of the AI that enables this? Also collapsing:
| Metric | Reduction |
|---|---|
| Inference costs (via model cascading, distillation, caching) | 60-90% |
Software that gets better without human maintenance--and does so at declining cost--is not becoming un-investable. It's becoming more investable than any previous generation of software.
Insull had to hire thousands to maintain his grid. The AI-native software company can maintain exponentially more customers with a fraction of the headcount.
Insull didn't just build power plants. He built political infrastructure.
He recognized that competing power franchises were wasteful--"two companies laying wires down the same alley" drove costs up, not down. So he advocated for something radical: regulated monopoly. He lobbied for state public utility commissions that would grant exclusive franchises in exchange for rate regulation.
His competitors thought he was mad. Why invite government oversight?
But Insull understood something they missed: the franchise was a moat. Once you had the regulatory bargain--the exclusive right to serve a territory in exchange for regulated rates--competitors couldn't just show up with a better generator and take your customers. The institutional infrastructure protected the physical infrastructure.
In software, platform lock-in is the new franchise.
Products built deeply into AWS, GCP, or Azure benefit from shared identity, prebuilt connectors, managed integrations, marketplace distribution, and deeply embedded billing relationships. Unwinding a SaaS deployment that sits inside a cloud ecosystem isn't a migration. It's a business transformation.
AI is pushing this lock-in higher up the stack:
- Enterprises are standardizing around specific model providers
- Agent frameworks are integrating deeply with cloud infrastructure
- Embedding stores, vector indices, and fine-tuned models are becoming system-level assets
Naval is not wrong about everything. Neither were Edison's loyalists.
He's right that feature-level differentiation is dying. If your moat is "we have a button that does X," you're in trouble--AI can generate that button overnight.
He's right that seat-based SaaS pricing is under pressure. When AI amplifies individual productivity 10x, charging per-seat makes less sense.
He's right that the speed of competition has accelerated. A two-person team can now build a credible v1 in weeks, not years.
And he's right that mediocre software companies--products that rely on UI skins, commodity features, and linear sales motions--will get squeezed.
But Edison's loyalists were also right that AC would standardize. They were right that anyone could buy a Westinghouse generator. They were right that the technology was commoditizing.
They were wrong that this made the electric business un-investable.
The companies that thrived weren't the ones clinging to proprietary generator technology. They were the ones building wires, installing meters, cultivating customer relationships, and securing franchises.
Software is the same. The generator--the code, the model, the features--was never the moat.
Samuel Insull died in a Paris metro station in 1938, clutching 30 francs. His empire had collapsed in the Depression. He'd been tried for fraud and acquitted, but his reputation never recovered. Franklin Roosevelt had campaigned on a promise "to get" the Insulls.
Yet his insight outlived him. The regulated utility model he pioneered still structures American infrastructure. The understanding that distribution beats technology still shapes how we think about network businesses. The recognition that the wire matters more than the generator still explains why platform companies capture more value than product companies.
The companies that will win the AI era are not the ones with the best models. They're the ones with:
- The densest networks (the wires)
- The deepest customer relationships (the meters)
- The most embedded workflows (the grid)
- The strongest platform lock-ins (the franchise)
And infrastructure, as Insull discovered on that winter evening in Brighton, is very hard to rip out of walls.
