If your business is bottlenecked by intelligence alone, you are on borrowed time. The bots aren't just coming; they are here, and they are rapidly commoditizing software scale, ecosystem lock-in, and engineering faster than you can describe them.
But there is an endgame. I wrote about this recently: the assets that actually get stronger as AI gets better. The core filter for defensibility hasn't changed, but the battlefield has decisively shifted from software back to the physical world. It comes down to one question: Is it hard to do, or is it hard to get?
If what you are building is genuinely hard to get—because it's bottlenecked by human behavior, by physics, by political will, or by sheer elapsed time—you are building something that lasts. Here is what real defensibility looks like when intelligence is virtually free.
The 5 Pillars of Real-World Defensibility
1. Compound, Ground-Truth Data
Look at Orchard AI. They mount cameras to farm equipment, tracking billions of fruit across millions of trees, spanning multiple growing seasons and geographies. Every day, every single pass through an orchard makes that dataset richer and the models smarter.
- The Moat: You simply cannot replicate this by training a model on public data. You have to put physical cameras in the real world and drive them through the same orchards for years. Time is the unbreakable barrier to entry.
2. Network Density and Liquidity
Every user that joins a true network makes the product exponentially more valuable for everyone else. DoorDash is the clearest example: every driver makes delivery faster, every restaurant adds choice, and every customer makes the economics work. An exchange fund like Cache proves this beyond marketplaces—a larger pool means better diversification, attracting more participants.
- The Moat: You can clone the app overnight with AI. But you cannot clone the drivers, the restaurants, the customers, and the delivery density across ten thousand cities. As AI makes it trivial to build alternatives, a hundred well-built clones will fight over scraps while the one with existing liquidity compounds.
3. Regulatory Permission
Governments move at the speed of politics, not technology. Take Anduril—they need procurement clearances and classified contracts to sell to the Department of Defense. A bank charter takes years. FDA approval takes years.
- The Moat: No amount of AI compresses the bureaucratic timeline. As AI capability scales, the stakes get higher, and the surface area of regulation expands. The specific forms of regulation will change, but the need for human permission isn't going anywhere.
4. Capital at Scale
This is the one almost everyone is underweighting. The endgame is physical. Chip fabs cost $20 billion. Nuclear plants cost $10 billion. Satellite constellations cost billions more. There's a reason Elon Musk is raising $75 billion and IPOing SpaceX even while predicting money might not matter in 15 years.
- The Moat: When the bottleneck shifts from software to atoms, the ability to finance and deploy capital at a massive scale becomes a defining advantage of the era. Capital access isn't just cash—it’s institutional trust, a decades-long track record, and relationships that take a lifetime to build.
5. Hard Physical Infrastructure
Factories. Power plants. Battery networks. Data centers. Base Power is deploying thousands of battery units across Texas homes while simultaneously building its own manufacturing facility. Every installed unit is a physical asset sitting in someone's yard, generating revenue.
- The Moat: Soon, you'll be able to design that entire system in a week using AI. But you cannot manufacture, install, and interconnect thousands of physical units in a week. Physics sets a hard floor on timelines that intelligence cannot break through. Whoever started building first has a lead that grows with every month a competitor hasn't broken ground.
The Scarcity Shift
Notice what didn't make the list: the things where time can be compressed.
- Workflow Embeddedness: It sounds durable until you realize the switching cost is just engineering time in disguise.
- Ecosystem Lock-in: It feels like a fortress until AI can rebuild integrations on the fly.
- Software Scale: Spreading engineering costs across millions of users stops mattering when the cost of engineering approaches zero.
These were moats against the scarcity of intelligence. That is the one form of scarcity we know is ending.
Every defensible asset requires years of real-world, elapsed time to accumulate. No amount of intelligence compresses that clock. Time that cannot be parallelized is the meta-moat. The companies that already hold these positions aren't just defensible; they are pulling further ahead every single day, because the head start is the moat.
The Unsolved Variables
There are two areas I'm still genuinely uncertain about:
- Trust and Liability: Does trust become its own moat when AI does more of the heavy lifting? Someone has to be accountable when things go wrong. The institution that bears that liability might become infinitely more valuable, not less.
- Human Attention: When the cost of creating content drops to zero, the brands that already hold our attention might have the most compounding advantage of all.
I suspect both are very real, though I haven't figured out whether they are their own distinct categories or simply byproducts of the five pillars above.
The bottom line: Look for the bottlenecks of time and atoms. If your moat is bottlenecked by years, human behavior, physics, or capital, you are building a legacy that lasts.
