Software Was Built on Seats. Agents Don't Sit Down.

For twenty years, the software business ran on one quiet assumption: that behind every login was a human. Vendors priced their products by the seat — the more people a company put in front of the software, the more the software earned. Usability, onboarding, adoption: the whole craft of enterprise software was built to get more humans touching more tools more often.

In early 2026, that assumption broke in public. In the first week of February, more than a trillion dollars in market value was erased from software stocks, according to Forrester, as investors absorbed a simple realization: autonomous agents do not log in. They do not hold licenses. They complete the work that seats used to represent, and they do it without sitting down. When the unit of labor stops being a person, the unit of pricing built around that person stops making sense.

The numbers behind the panic are not speculative. Gartner projects that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025, and expects at least 40% of enterprise software spend to move to usage-, agent-, or outcome-based models by 2030 — with the share tied to seats falling as it goes.

And among AI companies the shift is already underway, not just forecast:

But the pricing story is the surface. Treat it only as a pricing story and you will miss the thing actually changing.

From owning the tool to receiving the result

For a generation, buying software meant buying access to a capability you then had to operate yourself. You licensed the CRM and still hired the people to work it. You bought the marketing suite and still staffed the team to run the campaigns. The tool was the product. The labor was yours.

What agents collapse is exactly that gap. When the software can run the workflow end to end, the customer stops wanting access to a tool and starts wanting the outcome the tool was always a proxy for. Not a dashboard, but a decision made. Not a campaign builder, but qualified buyers on the calendar. Not a help desk, but the ticket resolved. The interface — the thing two decades of software was organized around — becomes the least interesting part of the transaction.

This is why the most exposed software is the kind whose entire value was merely providing an interface. The transition won't be frictionless: in complex or high-stakes sectors, human effort isn't eliminated, it shifts from daily operation to strategic oversight to manage compliance and security risk. Moving off per-seat also introduces new challenges — budget predictability, and the difficulty of attributing final business outcomes strictly to software. The layer that survives won't just remove the interface; it will bridge the gap between autonomous execution and necessary human supervision.

The companies that saw it early weren't contrarian. They were early.

It is tempting to read the February repricing as a shock. It is better read as the market catching up to a shift that was already underway in the places paying the closest attention to where work actually happens.

The businesses positioned for what comes next made a specific bet before it was obvious: that customers do not want more tools, they want fewer problems. They built their products to deliver a measurable result and priced themselves against that result, so their incentives and their customer's incentives pointed the same direction.

When the software does the work, you don't sell the software. You sell the work, done.

— Muan Group

That is the real lesson here. The software economy is not ending — it is being rebuilt around a different promise. For twenty years the question a buyer asked was how many of my people will use this. The question now is how much of my work will this finish. The companies with a good answer to the second question are about to take share from everyone still optimizing for the first.