The Moat Was the Cost of Building Software. Claude Code Just Mass-Produced a Bridge

THE NUMBER: $100 billion — The amount Jeff Bezos is reportedly raising to buy manufacturing companies and automate them with AI, per the Wall Street Journal. Yesterday we wrote about Travis Kalanick’s Atoms venture — $1 billion raised on a $15 billion valuation to bring AI to the physical world. Today one of the richest people on the planet walked into the same room at nearly 100x the scale. The atoms economy just got its first mega-fund.

A VC told Todd Saunders something this week that lit up X like a signal flare: “The moat in software was the cost of building software. And Claude Code just mass-produced a bridge.” The post hit 169,000 views in a day. Not because it was provocative. Because it was precise.

For thirty years, SaaS margins — Salesforce at 75% gross, ServiceNow north of 80% — existed because building a competitor required millions of dollars, a team of engineers, and years of iteration. The moat wasn’t the product. It was the cost of producing it. That moat is gone. Not eroding. Gone. And the bridge across it isn’t a single footpath — it’s what the Army Corps of Engineers built after the Germans blew the bridges in World War II. Mobile pontoon bridges. Dozens of them. Blow one up and three more appear downstream. That’s what AI coding tools just did to the software industry.

Meanwhile, the smartest operators of the last thirty years are converging on the same response: if the digital moat just collapsed, go build in atoms. Kalanick spent eight years stealth-building a physical AI company. Now Bezos is raising a $100 billion fund to buy and automate American factories. Two empire-builders, independently arriving at the same thesis: when software moats disappear, physical complexity becomes the only moat left.

The question isn’t whether AI reshapes both worlds. It’s who captures the value — and whether the returns are worth the bridge.

Fifty Thousand Companies, One to Three People, and a Question Nobody’s Asking

Saunders’ thesis is that AI coding tools don’t kill software value — they redistribute it. The SaaS boom produced a few dozen billionaires and a bunch of zero-sum competitors fighting over the same enterprise contracts. The AI SaaS era will mass-produce millionaires. Fewer ServiceTitans hitting $5 billion valuations. More like 50,000 companies doing $500K to $5M each, run by one to three people with deep domain expertise and enormous margins.

Harrison Chase — founder of LangChain — called this “harness engineering” at NVIDIA’s GTC this week. When you’re impressed by an AI product, he argued, you’re responding to the harness — the domain logic, the workflow design, the context management — not the underlying model. LangChain climbed from the bottom of Terminal Bench rankings to the top five by redesigning its harness without changing the model. Same engine. Better chassis. AI is cognitive leverage. It amplifies how far what you know can reach.

But here’s the edge that Saunders’ optimism undersells. When friction declines in any system, competition compresses returns. That’s not a fear — it’s arithmetic. If you were the sole gatekeeper on a large problem, you made billions. If you’re one of a thousand gatekeepers on discrete problems, you make millions. And over time, if switching costs are low enough and someone is willing to accept less, the market competes away profit altogether.

The real question nobody’s asking: when is your moat small enough that it’s not worth building a bridge to your castle? If I can build software and earn $100K a year doing it — is that enough? And will a developer in Ho Chi Minh City or Bangalore be happy to earn $10K serving the same niche, because $10K goes a long way in Vietnam? The barbell isn’t just big-generic-gets-crushed and small-niche-thrives. It’s that every niche eventually attracts someone willing to do it for less — unless you own the customer relationship, the proprietary data, or the workflow integration that creates real switching costs.

Cursor just shipped Composer 2, built on its own frontier model. A code editor company decided the model layer was too important to rent. That’s the playbook: own the stack or own the customer. Renting your position in between is a countdown.

The signal for builders: The window for AI-native vertical software is open now. The domain experts who move first — who build for lock-in from day one, not features — will capture the value. Everyone who follows will compete on price until the margins disappear.

Bezos Wants to Build the Factory. Literally.

The Kobeissi Letter broke it today: Jeff Bezos is in talks to raise $100 billion to buy manufacturing companies and use AI to automate them. Yesterday we covered Travis Kalanick’s Atoms — eight years stealth, operations in 30 countries, $1 billion raised on a $15 billion valuation. Kalanick described the thesis simply: manufacturing is the CPU, real estate is storage, logistics is the network. If you’re in the atoms world, you’ve decided you like hard things.

Now imagine someone worth $200 billion with a $500 million yacht who wants to be a startup CEO. Again. Someone who already built the logistics infrastructure that delivers packages to every doorstep in America and the cloud computing infrastructure that powers half the internet. That person just decided the next frontier isn’t space. It’s the factory floor.

Here’s the framing that matters: the wrong question is “will Bezos take over factories and fire everyone?” The right question is “can Bezos crack automated manufacturing to a degree that makes reshoring viable?”

Notice what he’s not doing. He’s not trying to make Chinese factories more efficient. He’s buying American manufacturing — and the implicit thesis is that AI can remove enough cost and complexity to make domestic production competitive again.

US manufacturing’s problem isn’t just that workers cost more than in Shenzhen. It’s the complexity tax. OSHA regulations. EPA compliance. Supply chain coordination across dozens of suppliers. Quality control loops requiring human judgment at every node. Insurance, liability, permitting. That overhead is what humans manage poorly at scale — and it’s what drove production offshore in the first place.

AI doesn’t just reduce labor cost. It radically simplifies the entire process. OSHA looks very different when there are no human bodies on the factory floor. Environmental compliance gets easier when you monitor every variable in real time. Supply chain coordination becomes trivial when an agent tracks every component and reroutes every shipment without a human in the loop. You’re not just removing the most expensive piece of the puzzle. You’re redesigning the puzzle.

This is the electric motor story. When electricity came to factories in the 1890s, the first instinct was to swap the central steam engine for a central electric motor and keep everything else the same. It took 30 to 40 years before factories were fundamentally redesigned around distributed motors — power at the point of use, flexible layouts, multi-story buildings impossible with belt-driven systems. The entire architecture had to change before the technology delivered its potential. In the AI age, that redesign won’t take 30 years. Maybe five. Because AI itself accelerates the redesign — you can simulate a thousand factory layouts in the time it used to take to blueprint one.

The logical entry point is the top of the complexity chain. Prove it in semiconductor fabrication, where precision is measured in nanometers. Prove it in aerospace, where a single defective part grounds a fleet. Prove it in defense, where the government will pay premium for domestic production with secure supply chains. Then cascade down. Consumer electronics. Automotive. Medical devices. Each step gets easier because the operating system — the AI-native factory management layer — is already proven.

The legacy question: Remember COVID. Remember the shortages. Remember waiting six months for a dishwasher. Remember discovering that the world’s largest economy couldn’t manufacture its own PPE. If Bezos builds an AI-powered manufacturing OS that makes reshoring viable — not by racing to the bottom on labor, but by eliminating the complexity that made offshoring necessary — the implications go far beyond portfolio returns. The cost of goods comes down. The reliance on adversaries comes down. And the vulnerability that COVID exposed gets patched. Maybe that’s the Bezos legacy. Not the everything store. Not the rockets. The guy who brought manufacturing home.

What This Means For You

The software moat and the manufacturing moat are collapsing for the same reason: AI eliminates the complexity premium that incumbents relied on. But the responses are different.

If you’re in software, the clock is ticking on generic value. Every SaaS company charging premium prices for features that can be rebuilt in a weekend needs to answer one question: what do we offer that the customer can’t build themselves? If the answer is “our brand” or “our integrations,” those are depreciating assets. The answer needs to be proprietary data, workflow lock-in, or domain expertise that compounds. Build for switching costs, not features.

If you’re in manufacturing, watch the complexity chain. Bezos won’t start with your factory. He’ll start where willingness to pay is highest — semiconductors, aerospace, defense. But every step he takes down the chain reduces the timeline for your industry. The question isn’t whether AI-automated manufacturing reaches your sector. It’s whether you’re operating it or competing against it.

If you’re allocating capital, the barbell is the trade. Short the generic middle — SaaS companies with commodity features and no switching costs, manufacturers with high labor costs and no automation roadmap. Long the extremes — domain-specific AI-native builders who know their vertical cold, and physical AI infrastructure plays that are years ahead.

Three Questions We Think You Should Be Asking Yourself

If a 60-year-old whose last programming language was Pascal can rebuild production software in a weekend, what exactly is your engineering team’s moat? It’s not the code. It was never the code. It’s the knowledge of what to build and for whom — the harness, not the engine. If your competitive advantage lives in your codebase rather than your customer relationships, you’re one Claude Code session away from a competitor.

When Bezos and Kalanick independently converge on the same thesis — physical AI, atoms over bits — are you paying attention to what the smartest operators in the room are telling you? These aren’t speculators. These are builders with track records measured in hundreds of billions of dollars of enterprise value created. When they both decide that the future is in factories, not software, the signal is deafening.

At what point is your niche too small to defend? The AI SaaS era will create 50,000 small software companies. But it will also create 50,000 competitors for each of them — including ones in countries where $10K a year is a great living. The winners won’t be the ones who build the best software. They’ll be the ones who build the deepest customer relationships before the next bridge appears.

“The moat in software was the cost of building software. And Claude Code just mass-produced a bridge.”
— A VC, via Todd Saunders

The Moat Was the Cost of Building Software. Claude Code Just Mass-Produced a Bridge

Fifty Thousand Companies, One to Three People, and a Question Nobody’s Asking

Bezos Wants to Build the Factory. Literally.

What This Means For You

Three Questions We Think You Should Be Asking Yourself

— Harry and Anthony

Sources

Past Briefings

Bill Gurley Says the AI Bubble Is About to Burst. Travis Kalanick’s Timing Says He’s Right.

Anthropic Is Winning the Product War. The $575 Billion Question Is Whether Anyone Can Afford to Keep Fighting

Chamath Says Your Portfolio Is Worth 75% Less Than You Think. Karpathy’s Data Suggests He’s Right.