Every conversation about AI and business seems to collapse into the same argument about jobs. It is the wrong argument for an investor. The useful question is not whether AI replaces people. It is which businesses AI quietly repositions from asset to liability, and which ones it turns into something worth considerably more. Those two piles are already forming, and the line between them is sharper than the headlines suggest. We only back the second pile.
We call the discipline unSaaSing: replacing rented software with infrastructure a company owns. But the underlying thesis is bigger than software procurement. It is a filter for which mid-market businesses compound in the AI era and which ones get repriced by it. This piece lays out the dividing line, the test we run before we back anything, and why owned infrastructure is the difference between AI as leverage and AI as a layoff notice.
The Dividing Line
AI is drawing a line straight through the middle market. On one side sit businesses whose core work was always, underneath the branding, just software moving data from one place to another. For those companies AI is a solvent. It dissolves the moat they charged for, and it does it fast, because the thing they sold is now a feature inside a general-purpose model.
On the other side sit businesses where a human still has to show up and something physical still has to happen. Freight that has to move. A patient who has to be seen. A crew that has to be on a roof by seven. Equipment that has to be serviced before it fails. For these companies AI is not a replacement for the work. It is leverage on top of it: it routes the trucks, clears the back office, drafts the quote, and keeps the pipeline warm, so the same operator does more without adding headcount.
The mistake investors make is treating AI as a uniform force. It is not. The same technology that guts a commodity software reseller makes a regional logistics operator meaningfully more valuable. Same tool, opposite outcome, and the outcome is decided entirely by what the business actually is underneath.
Where AI Creates Leverage
When we say AI is leverage rather than replacement, we mean something specific and unglamorous. The highest-payback uses of AI inside a durable business are almost never the flashy ones. They are the boring capacity multipliers that let a proven operation do more with the team it already has.
- Routing. AI dispatches, sequences, and optimizes the physical work. The people still drive the routes, run the crews, and see the patients. The margin comes from doing the same work with less waste in between.
- Back office. Agents clear the administrative drag that every physical business runs on: intake, scheduling, documentation, billing follow-up. This is capacity returned to the people who generate revenue.
- Demand. Automation answers inquiries in minutes instead of days, sends every follow-up on schedule, and reactivates customers nobody had time to call. It fills the pipeline without adding a single salaried seat.
None of these replace the operator. Each one makes the operator more productive, which is the entire point. A business AI can amplify gets structurally more valuable every quarter it runs. A business AI can only replace is on a countdown it cannot see.
The Resilience Test
Every business looks strong in a good year, which is exactly why a good year tells you almost nothing. Before we back a company, we run a resilience test designed to answer one question: what happens to this business when the technology changes again? We are not underwriting this quarter. We are underwriting the next shift, the one nobody has scheduled.
Three questions do most of the work:
- Demand. Is the need structural, or is it riding a trend that fades? Structural demand survives a downturn and a technology cycle. Trend demand evaporates the moment attention moves.
- Ownership. Do you own the data and the rails, or rent them from vendors who can change the terms? A business that owns its operating data can build on it. A business that rents it pays a tax on its own future.
- Dependence. Does the business run on one irreplaceable person, or on a system that holds when that person is out? Tribal knowledge is a single point of failure wearing an org chart.
Fragile and resilient look identical in the same good year. The difference only appears when the ground moves. One business absorbs the shift and keeps compounding. The other becomes the shift's casualty. The reason we test before we buy is simple: you can bolt on AI in a quarter, but you cannot retrofit durable demand or owned infrastructure after the fact. Resilience is the one moat that has to already be there.
Same Technology, Opposite Outcomes
The clearest place this shows up is the exit table. A leveraged operator running AI on infrastructure it owns widens its lead every quarter, and a buyer can see exactly why: the data is clean and owned, the automation transfers, the demand engine keeps running after the founder leaves. That business gets priced as an asset. A displaced business, one whose only moat was the software layer AI just commoditized, gets priced as a risk, because the buyer is underwriting a rebuild.
This is the same dynamic we described in the leaky bucket: rented workflows transfer as obligations, owned workflows transfer as assets. AI raises the stakes on both sides. It makes owned infrastructure more valuable, because it multiplies what that infrastructure can do, and it makes rented, commoditized capability worth even less, because the thing you were renting is now nearly free.
What UnSaaSed Actually Means
UnSaaSing is not an ideology and it is not a demand to write your own software. Some layers should stay rented: commodity utilities with low data gravity and honest pricing are usually fine as subscriptions. It is a portfolio decision made layer by layer, with numbers, about where ownership pays for itself.
What it means in practice is three things. Owned cores on mature open-source foundations, deployed on your own infrastructure and secured as company property. One data layer you control, so every AI use case starts with unrestricted access to your own operating data. And automation on rails you own, so the demand engine belongs to the business rather than to whichever vendor happens to see one slice of the funnel. Modern AI-assisted development is what changed the economics on all three, turning what used to be a multi-year build into a matter of weeks.
The businesses that win the next decade are not the ones with the biggest AI budgets. They are the ones that were already on the right side of the dividing line, and then put the leverage on infrastructure they own. That is what UnSaaSed means: a durable business, amplified by AI, compounding on rails it will never have to rent back.