Last month, we explored what appears to be a transition from software-as-a-service (“SaaS”) to a tokens-as-a-service (“TaaS”) model, and the Token Tax that comes along with it.

Meanwhile, the conversation happening in boardrooms, in PE offices, and in the back hallways where purchasers actually decide these things is about how to reframe the seat-based deal sitting at the foundation of the SaaS model. Homegrown, vibe-coded replacements for licensed software. Token-based intelligence bolt-ons. TaaS sidecars. All of it is the talk of the town, and yet curiously, the arguments almost always circle back to the same place, not tokens, but seats.

One hears the same question asked in one form or another, in the very rooms this author frequents: if AI really works, why are we still counting people?

It is an excellent question, and it may well become one of the defining questions of enterprise software over the next decade.

Some corners of the commentary have already given this moment a name, the SaaSpocalypse, apocalyptic in its confidence and, we suspect, premature in its diagnosis. Seat-based revenue is not ending. It is being renegotiated, and there is a considerable difference between the two.

AI BREAKS THE PROXY

Artificial intelligence introduces an awkward possibility for healthcare. Imagine a health system with hundreds of revenue cycle employees, and imagine AI successfully automates thirty percent of the repetitive work. The workflows remain. The integrations remain. The platform remains. The data remains. Only the labor changes.

Has the software become less valuable, or has it become more valuable, accomplishing the same work with fewer people? That distinction matters a great deal, because if the answer is “more valuable,” the seat was never really the product. It was simply the easiest way to measure value in a world where people performed nearly all of the work.

NOT ALL SEATS ERODE AT THE SAME RATE

A useful distinction has begun circulating among software investors, drawn between probabilistic and deterministic work. AI is, by nature, probabilistic. It predicts, ranks, drafts, and synthesizes, and in categories where a strong likelihood of being right is good enough, that is precisely what customers want. But other categories tolerate no such margin. A claim is either paid correctly or it is not. A control is either enforced or it is not. An audit trail either holds up under scrutiny or it does not.

The revenue cycle sits astride both worlds, and unevenly so. Coding, denial narrative, and the interpretive reasoning behind an appeal are probabilistic tasks, judgment calls made under uncertainty, which is exactly where an agent can begin doing real work. Payment posting, identity verification, and regulatory reporting are closer to deterministic, and here the margin for error narrows to nearly nothing. Compliance and audit obligations, HIPAA chief among them, do not bend to a model’s confidence interval. A determination is either defensible under scrutiny or it is not, and no health system will accept “usually correct” as a standard for anything a regulator might later ask it to explain.

This matters for how quickly seats actually erode. Where the work is probabilistic, the governed persona described above can plausibly absorb it. Where the work is deterministic and heavily regulated, the seat, and the human judgment and accountability behind it, will likely persist far longer than the more enthusiastic AI commentary suggests. The shift from SaaS to TaaS will not arrive evenly across the revenue cycle. It will arrive first wherever judgment, not enforcement, is the product.

THE PLATFORM WAS ALWAYS THE PRODUCT

This is where the conversation begins to change. Hospitals do not purchase enterprise software because they enjoy licensing thousands of users. They purchase it because they need an operating platform: clinical workflows, revenue cycle, scheduling, identity, integration, data, governance. The people were simply the interface to all of it.

AI is beginning to separate the platform from the labor. Once that separation is visible, software companies face an obvious question: if seats are no longer the best measure of value, what is?

THE GOVERNANCE GAP

This is also where the homegrown, vibe-coded replacement mentioned earlier tends to fall apart. It is one thing for a team to prompt its way to a working prototype over a weekend. It is quite another to run a regulated revenue cycle on software with no vendor behind it, no audit trail built for scrutiny, no one accountable when a regulator or a payer comes asking how a determination was made. Governance was never the most exciting part of enterprise software, and it was rarely what sold the deal. But it is very often what justifies the price of one, and it is precisely what a weekend’s worth of vibe coding does not yet provide. Until it does, the platform retains a form of value that has nothing to do with seats or tokens at all: someone to hold accountable, and a system built to prove it.

A CORRECTION, NOT A SUBSTITUTION

Much of the current commentary treats this as a simple substitution: seats decline, usage-based revenue rises to take their place. It is a tidy story, and there is real truth in it. Multiples on pure seat-based revenue have compressed noticeably, and consumption-based lines are growing faster almost everywhere they appear.

But that framing misses what we described three weeks ago as the governed persona, and it is worth restating plainly here. In practice, the seat rarely disappears. It is relicensed. The agent is issued its own credentials and its own place on the roster, much as a new employee would be, and the vendor collects a subscription for that seat much as before. What changes is the second line beneath it, the usage-based charge for the intelligence the agent consumes. Seats are not being replaced by tokens so much as joined by them. The invoice grows longer, not shorter.

There is already a word going around for a related but distinct habit, tokenmaxxing, employees and vendors alike treating raw token consumption as a proxy for productivity, burning compute because the meter is running rather than because the work demands it. We borrow its shape for a narrower purpose. Call it TaaSMaxxing: the vendor who keeps every seat on the roster, adds an agent persona for good measure, and then prices the tokens as though the seat were never a sunk cost to begin with. Not every hybrid pricing model deserves the label. But buyers would do well to know it when they see it.

THE NEXT UNIT OF VALUE

We do not pretend to know where the market ultimately settles. Perhaps enterprise licensing. Perhaps consumption. Perhaps transactions. Perhaps outcomes. Perhaps some hybrid no one has invented yet.

What seems clear is that every major technology transition eventually becomes an economic transition. Cloud computing changed deployment. SaaS changed licensing. AI appears poised to change how enterprise software measures value itself. That feels less like disruption and more like evolution.

LOOKING BEYOND THE SEAT

The next decade will likely be filled with conversations about pricing: seats, tokens, transactions, enterprise agreements, outcome-based contracts. Those are interesting discussions, but they are not the question that keeps returning to mind.

Once we stop counting people, what exactly are we buying? Because if software is no longer merely automating work, perhaps we should stop measuring users and start measuring intelligence.

Outcome-based pricing will tempt many as the honest answer, since paying for a result rather than a seat or a token sounds like the end of the argument. Part of its appeal is structural rather than sentimental: an outcome is a single, visible number, and it cannot hide the kind of opacity that a stacked seat-plus-token invoice can. Our own advice to buyers, for what it is worth, is to be wary of the hybrid model itself. A contract that preserves the seat and layers a metered token charge on top asks the buyer to monitor two moving parts at once, one of them largely invisible until the invoice arrives. Where a clean outcome can be defined and measured, it is very often the better bargain, not because it is more virtuous, but because it is harder to obscure.

Outcome-based pricing works cleanly enough where the result itself is clean, a claim paid, a form submitted, an eligibility check resolved. It works far less cleanly the moment the outcome is contested, delayed, or shared across several hands before it resolves, which happens to describe the most complex and most valuable corners of the revenue cycle rather than the simplest ones. Whether intelligence can be priced by outcome at all may depend entirely on what kind of intelligence is doing the work.

There is a further complication particular to healthcare, and it is worth naming before anyone treats outcome-based contracts as the tidy resolution to this whole debate. A cleaner claim rate or a lower denial rate is not an outcome that any vendor controls alone. Payers are deploying their own intelligence against the very same claims, tuned toward the opposite goal, catching overcoding, flagging fraud, tightening what counts as medical necessity. The claim does not resolve in a vacuum. It resolves inside a contest between two systems, each growing more capable, each aimed at a different definition of success. A denial rate that improves this quarter may simply reflect a payer that has not yet adjusted its own model, rather than a specialist agent that has permanently solved the problem. Pricing on outcomes amid that much simultaneous change is a fine aspiration. Treating it as settled would be premature.

Perhaps this is the more honest way to put the whole matter. Tokenmaxxing, whatever else it is, was always a symptom of counting the wrong thing, effort mistaken for progress, volume mistaken for value. The better question, valuemaxxing if we must name it, asks of every seat, every token, and every contract alike: not how much was consumed, but what was actually accomplished. Healthcare, with its adversarial payers and its unforgiving margins for error, may turn out to be the industry least able to fake an answer.

That is where our conversation goes next.

===================================

‍Larrian Martin is the Chief Information Officer at GoSB, a specialty healthcare RCM company, He was formerly Senior Vice President at Envision Healthcare and holds an Engineering Science degree from Penn State University, where he studied in the Artificial Hearts Lab.