For three decades you paid your software vendors a tax on the people you hired. You did not call it that. The invoice called it a license, a subscription, a seat. The accounting called it a SaaS line item. But the shape of the thing was always the same. Add a salesperson, add a Salesforce seat. Add a support rep, add a Zendesk seat. Add an engineer, add seats in Jira, GitHub, Slack, Notion, Figma, and a dozen tools you forgot you were paying for. Your software bill rose and fell with your headcount, because the seat counted humans, and the human was where the work happened.
That arrangement held because, for thirty years, humans were the only thing worth counting. Software created value by making a person more productive, so the person was the unit you metered. The seat was a proxy. A clean, countable, defensible proxy that showed up in the org chart and the budget and the admin console. Vendors priced against it because it was there, and because it grew when you grew.
In February 2026 the proxy broke in public, and roughly 285 billion dollars in software market value went with it.
The week the market did the math
On February 4, 2026, Anthropic shipped Claude Cowork along with Claude Opus 4.6, a model built to run sophisticated professional tasks and to spin up and coordinate whole teams of agents. Cowork reads your files, organizes your folders, drafts your documents, and over the following days picked up plugins tuned for sales, finance, marketing, and legal work. CNN Business reported a broad selloff across enterprise software as investors concluded that tools like Claude would do the work the software used to require a person to do.
The numbers were not gentle. The iShares Expanded Tech-Software ETF logged its worst two-day stretch since the 2008 financial crisis. Thomson Reuters fell 15.83 percent in a single session, its biggest one-day drop on record. LegalZoom sank almost 20 percent. FactSet dropped 10 percent. S&P Global, Moody's, and Nasdaq all slid. The trade press named it the SaaSpocalypse, which is an ugly word for a simple realization.
The market was not pricing in a better product. It was pricing in a broken meter.
Here is the chain the investors ran. A support team of fifty people used fifty seats of a support platform. Replace forty of them with an agent that resolves tickets, and the platform does not sell fifty seats anymore. It sells ten, or it sells something else. The revenue that rode on top of those forty humans does not migrate to the vendor automatically. It might flow to the model provider. It might stay with the customer who no longer pays for the work. What it does not do is keep flowing to a company that built its entire revenue model on counting people who are no longer there.
Software stocks recovered some ground on February 24 when Anthropic announced enterprise partnerships, integrating Cowork into Slack, Intuit, Docusign, LegalZoom, FactSet, and Gmail, which calmed the fear that the incumbents would simply be displaced. The bounce was real. The lesson underneath it did not change. The seat had stopped tracking value, and everyone in the room now knew it.
What the seat was actually counting
It helps to be precise about what per-seat pricing was, because the precision is where the strategy lives.
A seat is a license to one human identity. It assumes that value is created by a person sitting at a screen, and that more value means more people at more screens. Under that assumption, headcount is a perfect billing meter. It is observable, it is hard to fake, it grows with the customer's success, and it maps to a budget line every CFO already understands. For the entire SaaS era, software spend was a derivative of payroll spend. The two moved together. When you doubled your sales team, you doubled your CRM bill, and nobody thought that strange, because the software's value really did scale with the number of people using it.
Marc Benioff said the quiet part on Salesforce earnings calls. He has told investors that the better Agentforce gets, the fewer human agents a customer needs, which means fewer Service Cloud seats. He frames the agent product as digital labor. Read that again. The chief executive of the company that perfected per-seat SaaS is telling shareholders that his own best product shrinks the thing his old product was priced against. He is describing the severing of software value from human headcount, and he is trying to get ahead of it before it gets ahead of him.
Once an agent does the work, the human at the screen is gone, and the seat that human occupied is gone with them. The value did not disappear. The work still happens, faster and cheaper. What disappeared is the countable proxy the vendor used to bill for it. The output grew. The seat count fell. For the first time in the history of business software, those two numbers moved in opposite directions.
Salesforce changed its meter three times in eighteen months
Watch what a vendor does when its proxy breaks. It hunts for a new one.
Salesforce has shipped three different pricing models for Agentforce in roughly eighteen months. First it was two dollars per conversation, a clean per-interaction charge for customer chat. Then in May 2025 came Flex Credits, ten cents per action, sold in blocks of 500 dollars per 100,000 credits, so the meter ticked on every step the agent took rather than every conversation it held. Now the recommended motion for many deployments is a per-user license starting around 125 dollars a month, rebranded as digital labor, which quietly puts a seat back on the agent itself.
Three units of measure in eighteen months. Conversation, action, user. A company that knew what it was selling would not change the unit three times. The thrash is the tell. Salesforce is doing in public what every software vendor is doing in private, which is searching for a new thing to count now that counting humans no longer works. Each candidate unit is an attempt to re-anchor vendor revenue to something that still grows when the customer grows.
And every one of those candidate units points at you.
The new meters are a claim on your output
When the seat counted humans, the vendor's revenue grew when your team grew. That was a tax on hiring, and at least it was honest about what it taxed. The new meters tax something more intimate. They tax the work itself.
Look at the outcome model, which is the one the smartest vendors are running toward. Intercom prices its Fin agent at 99 cents per resolution. You pay when Fin answers a customer and the customer leaves satisfied, or when Fin completes a workflow that hands off cleanly. The pitch is irresistible on its face. Pay only when you get a result. Intercom reports that Fin now handles more than 80 percent of its support volume, resolves a million customer issues a week, and grew from one million to more than a hundred million dollars in annual recurring revenue on that 99-cent meter, backed by a performance guarantee of up to a million dollars if resolution targets miss. HubSpot followed the same road and in April 2026 cut its Customer Agent to 50 cents per resolved conversation. As of May 8, 2026, every Intercom support plan ships with Fin included, and the seat is now the side dish rather than the main course.
Sit with what outcome pricing actually is, underneath the buyer-friendly language. A 99-cent resolution fee is a royalty on your operations. The vendor is no longer selling you a tool you operate. The vendor is taking a cut of each unit of work your business completes. The better the agent performs, the more units it completes, and the more you pay. This is what an outcome fee really is. A revenue share you never sat down to negotiate, metered to each transaction, where the vendor gets paid most precisely at the moment your business is winning most.
This is a deeper entanglement than a license ever was. A seat capped the vendor's claim at your headcount. An outcome fee uncaps it and ties it to your volume. Scale your business tenfold without adding a single human, and the seat-based vendor earns nothing extra while the outcome-based vendor earns ten times more. The thing that used to protect you, the loose coupling between what software cost and what your business produced, is exactly the thing the new pricing is built to remove.
The margin math forces their hand
None of this is the vendors being greedy. This is the vendors being cornered, and the corner is real. The math behind it is unforgiving.
Bessemer Venture Partners published its AI pricing playbook in February 2026, and the central fact in it is that AI economics do not look like SaaS economics. Traditional software ran at 80 to 90 percent gross margin because the marginal cost of one more user was close to zero. AI products run at 50 to 60 percent margin, because every agent action burns inference, and inference costs real money on every call. Cost of goods sold, a number SaaS founders had the luxury of ignoring for twenty years, is back and it is large.
That changes the pricing problem from a preference into a survival constraint. A vendor cannot sell unlimited agent work under a flat per-seat fee, because each unit of work has a hard cost, and a heavy customer would torch the margin. So the vendor must meter usage to cover its compute bill, and then, having installed the meter, it would be foolish not to also capture the upside as your usage grows. Bessemer's data shows hybrid pricing, a base subscription plus usage or outcome tiers, jumping to 41 percent of AI vendors in 2026, up from 27 percent the year before. The vendor's compute bill becomes your variable bill. Their margin problem becomes your line item.
The renewal cycle is where this gets sharp. Many of the 2025 AI pilots hit their first real renewal in 2026, and Bessemer is blunt that pricing now has to reflect value delivered rather than value promised. Translation for the buyer. The cheap, generous, land-grab pricing of the pilot era is ending, and the renewal quote will be built on a meter designed to grow with everything your agents accomplish.
Who keeps the dividend
Strip away the pricing-page vocabulary and there is one number that matters. Call it the dividend. It is the gap between what a unit of work used to cost when a human did it and what it costs now that an agent does it. A support resolution that cost twelve dollars in loaded human time now costs a few cents of inference. That spread, multiplied across every task an agent now performs, is the entire economic prize of this moment.
The whole pricing fight, every conversation about seats and actions and outcomes and resolutions, is a fight over who keeps that dividend. There are only three places it can go. It can stay with you, the operator who deployed the agent. It can flow up to the model provider whose intelligence did the work. Or it can be captured by the application vendor sitting in the middle, whose meter decides how much of the spread it skims on the way through.
When you buy an agent priced per outcome, you are handing the dividend to the vendor by default. You rent your own efficiency gains back, one resolution at a time, at a price the vendor sets and re-sets at every renewal. You did the hard part. You redesigned the workflow, retrained the team, took on the risk, absorbed the change. And then you signed a contract that routes the reward to whoever owns the meter.
The vendors understand this perfectly. It is why Salesforce is willing to look indecisive with three pricing models, why Intercom built a hundred-million-dollar line on a 99-cent fee, why HubSpot is racing the price down to win the resolution before a competitor does. They are not fighting over features. They are fighting to be the layer that gets to define the unit and run the meter. Whoever defines the unit decides when the meter ticks, and whoever decides when the meter ticks owns the dividend.
Own the meter or rent your own gains
This is the part no procurement negotiation will solve for you, and it is the reason the right response is architecture, not purchasing.
The buyer who treats this as a vendor-selection problem will spend the next two years comparing per-resolution rates across vendors, congratulating themselves on shaving a dime off the unit price, and missing that the unit itself was chosen to extract the dividend regardless of the rate. The buyer who treats it as an architecture problem asks a different question. Where does the meter sit, and do I own it?
Owning the meter means owning the layer where your agents' work gets defined, counted, and valued. It means your own agent orchestration over your own systems of record, so that a resolution is something your business defines rather than something a vendor's pricing page defines. It means the accounting of what your agents did, what it was worth, and what it cost lives inside your walls, where the dividend stays yours. It means treating the foundation model as the metered input it actually is, a commodity you buy by the token from whichever provider is cheapest this quarter, while the value-defining layer on top of it is something you build and control.
This is the line between companies that will compound the AI dividend and companies that will hand it away. The first group is building the system that decides what a unit of their work is worth, because they understand that definition is where the margin lives. The second group is signing outcome contracts and discovering at renewal that they have given a vendor a permanent, growing, uncapped claim on the most productive part of their own operation.
Per-seat pricing was a tax on hiring, and at least a tax on hiring discouraged nothing you wanted to do less of. The new meters tax production, and a tax on production falls on exactly the thing you built the agents to do more of. You can negotiate a tax rate. You cannot negotiate your way out of a structure that takes a cut every time your business succeeds. You have to build past it.
The seat is not coming back. The question on the table for every operator right now is not which agent vendor to buy. It is whether the system that defines and meters your agents' work will belong to you or to someone selling you back your own efficiency at 99 cents a unit. That decision is being made in architecture diagrams this quarter, and most companies are making it by default, by signing whatever the pricing page offers.
Make it on purpose. Architecting the layer that owns your value capture is the opposite of a procurement task, and it is not something you can buy off a shelf, because the entire point is to stop being the party whose meter someone else controls. This is the work we do with operators who saw the February numbers and understood what they meant. Schedule a strategic consultation with us today.