Roughly two trillion dollars has come off software market capitalisation since the start of 2026. Gartner separately estimates that up to 234 billion dollars of enterprise application spending, around a fifth of the total, is exposed to the shift toward agents by 2030.
Those are public market numbers and it is tempting to file them under things that happen to other companies. The mechanism underneath them is not a market mood. It reaches any business that sells software by the seat, and it is already showing up in renewal conversations.
What the market actually repriced
Software valuations rested on a specific and reliable story. Customers adopt the product, headcount grows, seats grow with it, and revenue compounds without proportional cost. Net revenue retention above 100 percent came largely from existing customers buying more seats each year.
Agents break the link between headcount and value delivered. If a support team of forty handles the same volume with twenty five people and a set of agents, the seat count falls even though the work and the outcome are unchanged. The customer is not less successful. They are less numerous.
That is what got repriced. Not the quality of the software, and in many cases not even current revenue. The assumption that revenue grows with customer headcount stopped being safe, and every valuation built on that assumption had to be recalculated.
The uncomfortable version for smaller vendors
Public companies get analysts and headlines. A twelve person SaaS business with 300 customers experiences the same force with no coverage at all, and usually notices it later.
It shows up as renewals that arrive with a request to reduce seats rather than add them. As a churn number that looks stable while contract values quietly compress. As a sales cycle where the buyer asks what your product does that their own tooling could not now do internally.
That last question is new and it is not rhetorical. Roughly a third of organisations report skipping at least one software purchase this year because they could build the equivalent internally with agentic coding tools. Your competition now includes a customer's own engineering team on a quiet fortnight.
The vendors we work with who are least exposed are not the largest. They are the ones whose product is deeply tied into a workflow that would be painful to unpick, which turns out to matter more than scale.
We now ask a blunt question early in any product conversation: if a customer's team could do this themselves next year, what would still make them pay you. The answer is usually shorter than anyone expects and it is the thing worth building around.
Why the market moved before the revenue did
The striking part of the repricing is how little of it was driven by reported results. Many of the companies that lost the most value were still growing. What fell was the multiple applied to that growth.
Multiples encode a belief about the future. A business growing 30 percent with strong retention was worth a high multiple because the growth looked durable and the cost of serving each new seat was near zero. Both halves of that are now questioned: durability, because seats may shrink, and the cost, because agent inference is a real variable cost that scales with usage rather than sitting near zero.
That second point gets less attention and matters commercially. Traditional software had close to zero marginal cost per additional user. Software that runs models on the customer's behalf does not. Gross margins that sat comfortably above 80 percent for two decades are being recalculated with inference in them, and a business that prices flat while its costs scale with usage has an exposure that grows precisely when the product succeeds.
We covered the mechanics of moving off seat pricing separately in repricing when agents cut your seat count. This is the question underneath it: whether the market believes your revenue can survive your customers getting more efficient.
Your buyer now has a third option
Software purchases used to be a choice between vendors, or between a vendor and doing nothing. There is now a third column on the evaluation, and roughly a third of organisations chose it at least once this year: build it internally with agentic coding tools.
That option is not credible for everything, and it is credible for more than it was. A reporting layer, an internal dashboard, a workflow tool that moves records between two systems. These were bought because building them was not worth the engineering time. The engineering time changed.
What this does to your sales process is specific. Discovery calls now include a question about what makes your product hard to replicate, asked by someone who has already had an engineer estimate the build. Answering with feature lists loses that conversation, because features are what generation is good at. Answering with data, integrations, compliance posture and the cost of being wrong wins it.
What survives the repricing
Not everything is exposed. The products holding value share characteristics worth checking your own against.
They own data that cannot be trivially reproduced. A system of record accumulating years of transactions is difficult to replace regardless of how good code generation becomes, because the value is the history rather than the interface.
They sit in workflows involving compliance, audit or multi party coordination, where the cost of being wrong is high and buying something proven beats building something plausible.
They serve a specific industry deeply enough that the domain knowledge is the product. Vertical software has held up better than horizontal tooling for exactly this reason, which we covered in why vertical SaaS keeps winning.
The products under most pressure are horizontal, thin, and priced by the seat. A tool that is essentially a nice interface over a database, sold to any industry, charged per user, is describing what a competent team can now assemble internally.
What to do about it this quarter
Work out how much of your revenue is a headcount bet. Take your largest accounts and model what happens if their team shrinks 20 percent while their usage stays flat. If that materially damages your year, you have found your exposure and it is worth acting on before a renewal makes it urgent.
Then look at what your product does that is genuinely hard to reproduce. Be honest, and ask a customer rather than answering internally. The answers are usually about data, integrations that took years, or an edge case nobody would think to build until they hit it.
Finally, run a pricing experiment somewhere low risk. A new segment, a new region, or an add on module. Learning how your buyers respond to usage components on a small population is far cheaper than discovering it during a company wide migration.
If you are rethinking how your platform charges, our SaaS development team has done this migration with metering, caps and forecasting built in, which is the part that determines whether customers accept it.
Frequently Asked Questions
Is this just a stock market correction rather than a real shift?
Valuations always overshoot in both directions, so some of the two trillion is sentiment. The underlying change is not sentiment. Buyers are genuinely reducing seat counts, genuinely asking whether they could build alternatives, and genuinely comparing your pricing to vendors charging per outcome. Those show up in renewals whatever the market does next.
Should we abandon per seat pricing entirely?
Usually not. Seats are simple to understand, easy to forecast, and buyers are comfortable with them. The productive move for most vendors is hybrid: keep a base subscription for access and add a usage or outcome component for the work your product performs autonomously. That is where roughly half the industry has landed.
Our customers are not using agents yet. Does this affect us?
Not this year, and the exposure builds quietly. The question worth asking is whether your revenue depends on your customers' headcount growing. If it does, you are carrying a risk that is invisible until a renewal, and the cheapest time to address it is before any customer raises it.
How do we price an outcome we cannot always deliver?
Carefully, with a definition both sides accept before signing. A resolution that the customer reopens is not a resolution. Vendors doing this well define the outcome narrowly, measure it transparently, and let customers audit the count. Vague outcome definitions generate disputes that cost more than the revenue.
Does this mean building custom software is now cheaper than buying?
For some categories it has genuinely moved. Generation speed improved substantially, and a third of organisations acted on that this year. What did not improve is deciding what to build, maintaining it for five years, and owning the consequences when it breaks. The build decision is now more attractive and still carries obligations that a subscription does not.
What is the single most exposed kind of product?
Horizontal tooling, thin functionality, per seat pricing, no proprietary data. If a customer's own engineers could assemble something adequate in a fortnight with modern tooling, the product is competing against that possibility whether or not anyone attempts it.
How quickly should we change pricing?
Faster than feels comfortable, more slowly than a single decision. Start with new customers or a new module this quarter so you learn how buyers respond, then plan the migration of existing accounts over a year with grandfathering. Vendors that wait until a renewal forces the conversation negotiate from a much weaker position.
Talk to our team if your revenue is a headcount bet and you would rather find that out now than at renewal.