The Asset Durability Test

I spend my time evaluating what is worth building for the long term — as an advisor, as an investor, and as an operator. That question has changed.

AI is collapsing the cost of building things. Teams that once required dozens of people can now be run by a handful. Entire products can be developed in weeks that used to take months.

When execution becomes cheap, a different question matters more:

What cannot be easily replicated?

The series I have been writing — The Era of Human Agency — is about what that means for roles, organizations, and the businesses that employ them.

This piece asks the investment version of that question:

As AI commoditizes execution, what remains durable?

Not what is popular, or what is growing, or what is well-funded — but what is hard to replicate, resistant to abstraction, and worth defending over a long time horizon and thus has long term cash flows to value?

The Three Durable Assets

In the AI era, durable businesses increasingly come from three sources.

I think of these as the durable assets — the advantages that remain difficult to replicate even when execution becomes cheap:

1. Physical presence — assets that must exist in the real world
2. Operational data — information generated through doing the work
3. Niche context — domain expertise that resists abstraction

The rest of this essay explores why these assets matter — and where they are starting to converge.

Durability Is Not a Moat

Start with a distinction.

Durability is not the same as a moat.

Moats are competitive advantages — network effects, switching costs, economies of scale. Since everyone listens to Acquired now, this is where you can insert a comment about Helmer and his 7 forces.

AI weakens several of them.

Scale advantages shrink when a five-person team can match the output of fifty. Switching costs drop when AI handles data migration and onboarding. Network effects hold, but only for system that have already achieved density.

Thus, durability is narrower and harder.

A durable advantage solves a persistent problem using something that is difficult to replicate.

Both conditions must hold. Solving a persistent problem with replicable tools is a commodity business. Holding a unique asset that does not solve a real problem is a science project.

When execution becomes cheap, the only advantages that survive are the ones competitors cannot execute into existence.

Side note: There is a legitimate alternative. You can build for speed, volume, and disposability — the "fast fashion" model and AI has made this faster and more competitive than ever. Low cost, high velocity, optimized for attention rather than retention. Political campaigns. Trend-driven consumer brands. Content arbitrage. Consulting and other services.

AI makes this model more powerful too. It lowers the cost of producing at volume and responding in real time to almost zero. If that is your game, AI is an accelerant and you do not need a durability thesis, you need a velocity thesis. And you do need to understand the game you are playing and build for scale economies or keep investment extremely lean.

That is a valid strategy. It is not mine.

I invest in and advise businesses built to compound over years, not capture attention over weeks. Everything that follows reflects that bias.

1. Physical Presence

AI cannot show up.

It cannot occupy a location, maintain a machine, inspect a building, or shake a hand.

Businesses built around physical assets such as real estate, equipment, infrastructure, logistics that retain a durability that many software businesses do not. The limit here is atoms and there are only so many of them; or as my eight grade history teacher put it "Land = Wealth = Power" (s/o to Coach Hall for that one living rent free in my head for 30 years).

As the digital cost curve flattens, the relative value of what cannot be digitized increases.

A company like Copart illustrates this dynamic. Its salvage vehicle auction marketplace is valuable, but the real advantage is the network of vehicle yards and logistics infrastructure built over decades. A competitor can replicate the software. Replicating the physical footprint — land, transport routes, and insurance relationships — is much harder.

A stealth investment of mine is building a new take on durability asset, enabled by purpose built AI, for one of the largest physical asset markets out there, homes. I expect many more players across these markets to emerge.

The irony of the AI era may be that the most defensible businesses are the least digital. This inverts the venture theses of the last 30+ years of software - backed to the days of real risk and real assets. Not because software does not matter — it does, enormously, as an amplifier — but because the physical layer is the one layer that cannot be replicated by a competitor with a credit card and an API key.

2. Operational Data From "Real Work"

The most defensible data is not scraped, licensed, or aggregated from public sources.It is generated through direct interaction with customers, operations, or environments — and it compounds over time.

In fact, most enterprise data that is not publicly available is created through real-world activity: machines running, workers performing tasks, systems moving goods, and decisions being made.

Even digital infrastructure businesses follow this pattern. Stripe's payments platform processes billions of transactions across its network, generating continuous feedback on fraud patterns, payment success rates, and merchant behavior. That data improves the system over time because it is generated through real economic activity, not scraped from public sources. In that sense, Stripe functions as digital infrastructure — the system becomes more valuable because it sits inside the flow of real-world activity.

I'm partnering right now with a stealth startup working on this thesis - specifically that we have new ways to understand this data at scale and make better decisions. More on that company in the next few months as they go to market.

The data is not the product. The data is the byproduct of doing the work — and the accumulation of it becomes a structural advantage that AI amplifies rather than replaces.

This is the pattern I spend most of my time on — businesses that generate proprietary data through the execution of their core work, then use that data to make their operations and their clients' operations structurally better over time. It is the core of my work in the workforce and AI infrastructure space.

The most durable data is not harvested. It is handmade. Generated one task at a time through real work by humans, agents and software.

A new entrant can match your tools overnight. They cannot match your accumulated operational knowledge overnight.

3. Niche Context That Resists Abstraction

Some markets are small enough, specific enough, or regulated enough that generalist AI tools cannot serve them well.

A compliance workflow in specialty insurance. A procurement process for rare materials. A credentialing system for a licensed profession. A quality control protocol for a manufacturer with unusual tolerances.

These problems do not scale horizontally. They go deep. Depth creates durability that breadth does not.

Companies like Veeva Systems illustrate how powerful this dynamic can be. Veeva dominates software for pharmaceutical companies not because CRM technology is difficult, but because the regulatory and operational context of drug development is extraordinarily specific.

Generalist AI is powerful precisely because it is general. But generality becomes a weakness when the problem requires context that only comes from years of operating inside a specific domain.

One of my favorite investments in this space is Anuvi, focused on rebuilding Cobra for the modern era. Matt, the founder/CEO and his team have targeted specific problems with deterministic and marketplace needs, and are out executing everyone to solve that. The product is set up to be a great system of record and marketplace while AI agents run around and enroll people.

The key for niche businesses is capitalizing leanly enough to stay small and focused. The moment a niche business takes on capital that demands platform-scale returns, it is forced to dilute the depth that made it durable in the first place.

Robotics: One Spot where these Assets Converge

One place these durable assets are starting to converge is robotics — and it is a space I am actively paying attention to as an investor.

Robotic systems operate in the physical world, which means they combine physical presence with the generation of proprietary operational data.

Every robot deployed in a warehouse, factory, or hospital continuously generates information about how real environments behave. Over time that operational dataset compounds. And because robotics deployments are embedded inside specific industries — logistics, manufacturing, agriculture — they accumulate deep domain context as well.

The hardware creates presence. The operations generate data. The industry context creates specialization.

Over time, deployment itself becomes an advantage — because systems operating in the real world continuously generate the data and context that competitors lack.

Many durable marketplaces emerge from this same pattern. When supply is tied to physical systems, operational data, or deep expertise, the marketplace becomes the coordination layer around a scarce asset.

The Durable Asset Test

If you are building a business in this era, the Durable Asset Test is simple:

Could a well-funded competitor with access to the same AI tools replicate what you have in twelve months?

If yes, you are likely in a commodity business — possibly a profitable one, but not one I would invest in. The economics may work today. They will not compound.

If no, ask why. The reasons usually trace back to the durable assets: physical presence, operational data, niche context.

Durable companies of the next horizon will own at least one of these assets. The strongest will combine two. The rare ones will build systems where all three reinforce each other.

In practice, this means many of the most durable AI companies of the next decade will not look like traditional software businesses at all. They will be embedded inside real-world systems where physical operations continuously generate proprietary data and domain expertise over time.

All that said, I think that economics is only part of the story. This explains why companies have models that are durable, all other things being equal. And, in my opinion, there is another dimension of durability — one less about tangible assets and far more human. That is the subject of the next piece.

Final CTA: If you are building something durable — particularly at the intersection of AI, physical operations, and workforce infrastructure — I am interested in the conversation, whether that means advisory, investment, or both. I'm actively writing small angel checks in this space. Please reach out on LI, Twitter or via email.