When Seats Die, Your Quoting Engine Has to Grow Up

Earlier this year, roughly $285 billion in SaaS market value evaporated in about 48 hours. The press called it the SAASpocalypse. The trigger was simple to state. Investors decided that if AI agents can do the work of ten people, companies stop paying for ten seats. Jason Lemkin put it in the sentence everyone has been repeating since. If 10 agents do the work of 100 reps, you need 10 Salesforce seats, not 100.

It is tempting to read that as a story about software vendors and their stock prices. It is bigger than that. The seat, the per-user license, the priced-by-headcount model, has been the default unit of value for a generation of SaaS selling. What the market just repriced is the assumption that the unit holds. And once buyers stop accepting “price per user” as the natural way to buy, that pressure does not stay inside the software industry. It moves to everyone who sells anything.

Your buyers are already there. They want to pay for outcomes, for usage, for what they consume, for results rather than access. That shift was underway before agents arrived. Agents just made it impossible to ignore.

Which raises a question that has nothing to do with whether your CRM vendor survives the year, and everything to do with how you sell. When the unit of value stops being a simple count, something else has to become the thing you price, quote, and stand behind. And that something lives in CPQ.

The pricing model is the real story

Aaron Levie has been making the durable version of the agent argument. His framing is that the revenue model slides along a continuum. Some value still gets sold as seats, and a growing share gets monetized through consumption as agents and automation do more of the work. You can already see the blend in newer products that pair a per-user component with usage billing on top.

Now push that down to your own deals. The future of how you price is not “seats die, consumption wins.” It is hybrid, and it is different for every customer. One account is a base subscription plus metered usage above a threshold. Another is a flat platform fee with outcome-based bonuses. Another is tiered consumption with volume breaks and a committed minimum. The pricing logic that used to be “quantity times rate” becomes a branching set of rules that change by customer, by product, by term, by month.

That is a CPQ problem, and it always was. The era of simple unit pricing let a lot of companies avoid admitting how primitive their quoting logic actually was. When a quote was quantity times a number, you could survive on a spreadsheet and a seasoned rep’s judgment. Consumption, outcome, and hybrid models do not forgive that. They demand pricing, eligibility, and approval logic that is genuinely codified, versioned, and reliable, because the deal structures are too varied and too conditional to live in anyone’s head.

So the first uncomfortable question the SAASpocalypse raises is not “will agents take our jobs.” It is “can our quoting engine even express the models our customers now expect to buy.” For a lot of companies, the honest answer is not yet.

The second problem: agents have to reach the logic

There is a second question behind the first, and it decides whether the agent automation everyone is excited about actually works in your revenue motion.

If agents are going to help run quoting, checking eligibility, applying the right consumption tier, structuring an outcome-based deal, routing approvals, they have to reach your pricing logic directly. Not by clicking through a screen. By asking the engine a question and getting a deterministic answer.

This is where Levie keeps planting a flag, and his framing is closer to the headless argument than he probably intends. His point is that mission-critical processes need to be defined in deterministic business logic, precisely because the risk of an agent improvising something wrong on any given day is real. He has watched agents leak data and damage production systems. The lesson is not “avoid agents.” It is “put the business logic in a deterministic system the agent calls, rather than in something it has to improvise against.”

That is the entire case for headless CPQ, made by someone with a public-company balance sheet behind him.

For most of the last decade, the quality of a CPQ implementation was judged by what the user saw. Clean guided selling. Smart product pickers. Approval screens that surfaced the right exception at the right moment. The interface was the product. Configuration meant shaping screens until they fit how sellers worked.

In that world, pricing rules, eligibility logic, and approval thresholds end up living inside the presentation layer. An agent can only reach them by impersonating a user. Which screen, which sequence, which field fires which rule, which section renders only after a particular value is chosen. The logic is not addressable. It is performed. And anything performed through a UI inherits every fragility of that UI, including the silent rule that only fires when a human happens to click in the right order.

You cannot build reliable consumption pricing, and you certainly cannot let an agent operate it, on top of logic that is only reachable by pretending to be a person.

Headless is the precondition, not the upgrade

Headless usually gets sold as modernization. Decouple the front end, gain flexibility, support more channels. All true, and all beside the point now.

In a world where deals are priced by consumption and outcomes, and where agents help assemble them, headless stops being a preference and becomes the precondition for the thing you actually need. An agent has to ask your quoting engine direct questions and get the same answer every time. Is this configuration valid? What is the price at this volume on this tier with these terms? Does this require approval, and from whom? Those are API calls. They have to return identical answers regardless of what screen exists, regardless of click order, regardless of whether a human is anywhere near the transaction.

That only works when the logic lives below the interface, in a service layer the UI also happens to call. When the engine is headless by design, the human screen and the agent are simply two clients of the same logic. They get the same answer because they hit the same endpoint. The agent is not impersonating anyone. It is a first-class consumer of the system, exactly like the screen.

This is the line that separates a platform that can actually support agents from one with a conversational front end bolted onto UI-bound logic and called AI.

What this means if you are rethinking how you sell

The question to ask is not “does our platform have AI features.” Everyone says yes. Two better questions, both about your own revenue motion:

First, can our quoting engine express the way our customers now want to buy. If your pricing logic cannot represent metered usage, tiers, outcomes, and hybrids as actual rules rather than manual workarounds, you are not ready for the models the market is already moving toward.

Second, can an agent reach that full pricing and quoting logic without going through a screen. If the answer involves any version of “the assistant drives the UI,” your logic and your interface are the same thing, and you will fight that system the moment you try to automate beyond a scripted demo.

Levie is also right about the scale of the work ahead. He has called the agent transition a services wave larger than the move to cloud, because agents rewire the business process itself rather than just changing how software is delivered. That rewiring is the real work in front of revenue teams. Excavating the pricing rules that currently live in spreadsheets and tribal knowledge, codifying them, and exposing them as logic an agent can stand on.

The seat era let everyone judge CPQ by what they could see. The era that follows will judge it by what it can price, and by what an agent can call. When seats die, the quoting engine cannot stay a pretty front end with a spreadsheet behind it. It has to grow up.

Build the engine an agent can talk to. The rest follows.