Every few years someone declares that software is eating the world. It mostly ate the paperwork. The systems of record we built over the last four decades capture, route and report on work, but people still do the work itself. That is why enterprises spend roughly ten times as much on labor and services as on software.
That ratio is starting to change, and the change explains almost everything about where we invest.
Two ways to sell AI
There are two ways to bring AI into an enterprise. The first is to add it to an existing product: a summarize button, a drafting assistant, a chat panel. It’s useful, and incumbents will ship it everywhere. But it is still priced per seat, still sized to the software budget, and the human still owns the outcome.
The second is to rebuild the workflow around the agent. The product takes a claim, a prior authorization or a security alert and carries it to resolution, handing off to people only for judgment calls and exceptions. The customer pays for completed work. The budget it competes for is the operations budget, not the IT budget.
A feature makes a person faster. A workflow replacement changes what the company needs people for.
Why durability is the hard part
The obvious objection is that foundation models keep improving. If a general-purpose model can process a claim next year, why would anyone pay a specialist company to do it?
For many products, they won’t. Anything that is mostly a prompt and an interface will be absorbed by the platforms. The companies that last build on what a general model can’t download: data produced by doing the work thousands of times, integrations that took years of customer trust to earn, evaluation suites that encode domain judgment, and the liability and compliance posture that regulated buyers require.
Those advantages compound. Each completed task produces corrections and outcomes that make the next task more accurate. Accuracy builds trust, trust expands scope, and scope produces more data. In diligence, that loop is what we try hardest to measure.
What we look for at Series A and B
- Daily reliance. The product sits in the critical path. Usage is daily, not weekly or occasional.
- Improving unit economics. As automation rates rise, gross margin rises with them.
- Expansion from outcomes. Customers buy more because the product did the work, not because more seats were added.
- A clear answer to “why not the model?” Founders who can say exactly which part of the value a frontier lab will never prioritize.
This is the shift we’re built around. If you’re building on the right side of it, we’d like to meet you.