Every prior downturn in enterprise tech had an external trigger. The dot-com crash had a valuation bubble. 2008 had a credit crisis. The 2022 pullback had interest rate hikes repricing growth at any cost. Each time, the industry could point outward — at the market, the Fed, the macro environment — and treat the correction as something that happened to enterprise software, not something caused by it. The next correction won’t have that excuse. It’s going to be caused by an industry refusing to admit that its core product just changed underneath it.
The two assumptions everything was priced on
Enterprise software valuations over the past two decades have rested on two assumptions so foundational that almost nobody states them out loud anymore. First: revenue scales with the customer’s headcount, because pricing is per seat and headcount reliably grows. Second: switching costs are high, because software gets deeply embedded into a company’s workflows and ripping it out is expensive and risky. Together, those two assumptions are what justify multiples built on predictable, compounding renewal revenue.
Agentic AI breaks both assumptions at the same time, which is what makes this moment different from past corrections rather than a continuation of normal cyclical pressure. It shrinks the customer’s headcount-linked spend, because agents absorb work that used to require adding people. And it lowers switching costs, because the orchestration layer that does the real decision-making increasingly sits above the legacy systems of record rather than locked inside one of them — which means the legacy system underneath becomes easier, not harder, to eventually replace.
Why “we added AI” doesn’t fix the math
The standard incumbent response — adding an AI assistant or copilot to an existing seat-priced product — looks like adaptation but doesn’t address either broken assumption. The product still bills per seat. The customer’s headcount is still flattening. The assistant might even accelerate the very efficiency that’s shrinking the bill. Bolting AI onto a pricing model built for a world where software needed more people to use it doesn’t fix the unit economics. It delays the repricing by a renewal cycle or two, at the cost of looking responsive in an earnings call.
This is why the next correction in enterprise software valuations, when it comes, is going to look self-inflicted in hindsight. The market isn’t going to be blindsided by an external shock. It’s going to slowly notice that growth-per-seat stopped meaning growth-in-spend, across an entire category that priced itself as if that equivalence would hold forever.
What actually survives the repricing
The platforms positioned to come out ahead aren’t necessarily the most established ones. They’re the ones built agent-native from the start, pricing around completed work and outcomes instead of provisioned logins — a model that grows with the customer’s actual output rather than shrinking as the customer gets more efficient. That’s a structurally different bet than defending a legacy pricing model with a new feature.
Nobody is going to ring a bell when this correction starts. It’s going to look, for a while, like a normal soft patch in enterprise IT spending — until the renewal data makes clear that the soft patch was never going to recover, because the assumption underneath the entire category’s valuation quietly stopped being true.
