When Work Costs Sixty-Two Cents: The Metering Wars Quietly Repricing the Soul of Software

The first time I watched a machine resolve a customer support ticket for sixty-two cents, I did not feel the thrill the vendors promised. I felt grief — the specific, quiet grief of watching a profession lose its arithmetic. I have spent twenty-six years inside this industry, long enough to remember when a software license was sold by the human seat, a unit of billing that assumed a person was sitting somewhere, breathing, deciding. That assumption carried a moral weight nobody named. It said: value comes from people. When the same ticket that once cost seven dollars and forty cents to resolve falls to sixty-two cents, the price of the work changes. What changes more slowly — and more dangerously — is the story we tell ourselves about why anyone deserves to be paid at all. This is not a story about efficiency. It is a story about who gets to define a unit of value, and what happens to everyone else when that definition moves. For those who have not been following the enterprise software beat, "Agentic Commerce" is the industry's new catechism. The premise is simple enough: instead of selling a human a seat at a monthly table, software vendors now sell access to autonomous AI agents that perform the work directly — answering the customer, closing the deal, filing the report. Gartner estimates that $234 billion in enterprise software spending will be exposed to what it calls "agentic arbitrage" by 2030. George Brocklehurst, a managing vice president at the firm, framed it with unusual candor: agentic AI, he said, is changing the economics of software itself. He is right, but not in the way the vendors want you to hear it. The interesting question is not whether agents will do the work. They already do. The interesting question is how anyone plans to bill for it — and that question has fractured into four competing architectures, each with its own philosophy of value. There is per-action billing, exemplified by Salesforce's Agentforce at ten cents per action and fifteen cents per voice interaction. There is per-session or per-resolution billing, where HubSpot charges fifty cents and Zendesk between one-fifty and two dollars for each conversation resolved. There is per-outcome billing, still largely a promise, championed by startups like Ema. There is the hybrid model, where Monday.com adds AI credits on top of seats. And then there is Meta, which simply prices by the token — two dollars per million — undercutting the philosophical debate entirely with a price tag that is forty to fifty times cheaper. Four architectures. Four different answers to a question the industry cannot agree on: what, exactly, are we selling? Here is where the technical analysis begins, and where I want to be precise, because the sloppiness of the public conversation is doing real harm. Strip away the marketing and the "Agentic Commerce" debate is not a debate about pricing. It is a debate about three distinct technical layers that the vendors are deliberately collapsing into one: metering, settlement, and verification. Each has its own difficulty, and only one of them is genuinely unsolved. The vendors are betting you will not notice which. Metering is the easiest. It is the act of counting — actions, tokens, sessions, minutes. A token is a beautifully countable thing. It has no ambiguity, no edge cases, no room for a customer to argue. This is why Meta's token pricing is so strategically elegant: it is a price that cannot be disputed, only accepted or refused. Metering granularity and settlement automation are almost perfectly correlated. The finer the unit, the more the billing can run itself. Settlement is the second layer, and it is where the plumbing already exists. Stripe's settlement activity has been climbing vertically, and the older idea of a "Chain of Trust" — a verifiable pipeline through which autonomous entities transfer value to one another — has quietly moved from white paper to production. This layer is unglamorous, and it is winning. Whatever pricing model prevails, a payment must clear. The shovel-sellers do not care which gold rush wins. Verification is the third layer, and it is the one nobody wants to talk about, because it is the one that is genuinely hard. Consider what "per-outcome" billing actually requires. Someone must confirm, objectively and without dispute, that a desired outcome occurred. Did the agent close the sale, or did the sale close itself? Did the resolution satisfy the customer, or merely terminate the conversation? When the vendor both defines the outcome and grades it, we have handed the referee's whistle to the player. I have seen this movie before. In 2020, I led a governance working group at MakerDAO and reviewed more than five hundred voting proposals. What I found, and eventually wrote about in an essay called "The Quiet Collapse of Equity in Code," was that algorithmic neutrality is often a costume for whoever wrote the parameters. The math looked objective. The people it favored were not. This is the trap per-outcome billing is walking into. And it is why I find the industry's own confession so telling. One analysis put it bluntly: the industry is trapped between its desire for value-aligned pricing and its need for measurable units of work. That is the whole problem in a single sentence. The closer you get to measuring real value, the harder it becomes to verify and settle objectively. The closer you get to measurable units, the further you drift from value. You cannot have both. Every one of the four architectures is a different compromise point on a single, unavoidable curve. Here is what the vendors will not say out loud: the missing verification layer is the only place where this entire conversation touches cryptography. If per-outcome billing is ever to be more than a marketing promise, it will need exactly what blockchains were built to provide — objectively verifiable attestations, tamper-resistant settlement, and oracles that can confirm an external event without a central party adjudicating it. This is the genuine, unclaimed Web3 niche inside the agentic commerce story. And it is the part the industry has barely begun to design. Which brings me to the financial fact that overshadows every pricing debate: the gross margin. Traditional SaaS ran at eighty to ninety percent gross margin. AI-native businesses are running at fifty to sixty. That twenty-to-forty-point gap is not a rounding error. It is a structural repricing of an entire asset class. A per-outcome model that cannot pull margins back above seventy percent is not a revolution. It is a dignified decline. I keep returning to the economics of that single ticket. Human resolution: seven dollars and forty cents. AI resolution: sixty-two cents. The savings are six dollars and seventy-eight cents per ticket — a ninety-two percent cost reduction. That is the prize. And here is the question that should be keeping every enterprise software executive awake: who keeps the six dollars and seventy-eight cents? The customer, who now expects the discount? The vendor, who needs it to survive the margin collapse? Or the model provider upstream, who supplies the intelligence and can set whatever toll it likes? Whoever controls the metering standard controls the answer. That is why Salesforce is hedging across all four models simultaneously — not out of confidence, but out of the desperate hope that it can help define the standard before someone else does. To see where the value actually lands, you have to map the chain, and the map is unforgiving. Upstream sits the model layer — Meta and its peers — holding the strongest bargaining position because it owns the cheapest intelligence. Downstream sit the enterprise customers, whose power is rising precisely because AI hands them leverage they never had: the credible threat to replace human labor. And squeezed in the middle sits the software layer, the Salesforces and SAPs of the world, caught between a supplier that can set the toll and a customer that can walk away. This is a three-way squeeze, and the middle is bleeding. When Monday.com quietly abandons pure seat-based recurring revenue for the volatility of AI credits, it is not innovating. It is downgrading the quality of its own earnings, trading a predictable annuity for a consumption bet it cannot control. The market has not fully priced that trade. I should name something uncomfortable about how this story is being told. This is, at its core, an enterprise software narrative. It has almost nothing to do with on-chain cryptography, token economics, or consensus. The "blockchain" label is bolted on because payments settle somewhere, and payments have always settled somewhere. I say this not to dismiss the topic but to sharpen it: the genuine crypto relevance is narrow, specific, and — precisely because it is narrow — more valuable than the vague framing suggests. It lives entirely in that unbuilt verification layer. Regulation, for once, is not the headline risk — but it is waiting in the wings, and it is stranger than the compliance manuals anticipate. The moment autonomous agents transfer value through stablecoin rails, the ordinary machinery of AML, KYC, and money-transmitter licensing awakens. And beneath that lies a question no jurisdiction has answered: can an autonomous agent legally hold assets, enter contracts, or bear tax liability? We are building the economy of autonomous entities faster than we are building the law that would govern them. If Meta's forty-to-fifty-times price advantage is ever used to crush competitors rather than to serve customers, the more immediate regulatory threat is not crypto law at all. It is antitrust. The narrative you are being sold is that per-outcome billing is a customer-friendly evolution — you pay only for value received. Read the language of the people selling it and a different picture emerges. SAP's CEO called the shift a "unique opportunity to reset pricing." That is not the vocabulary of customer advocacy. That is the vocabulary of a company trying to escape a collapsing margin by re-anchoring its price to something other than cost. Per-outcome billing is a margin defense dressed as a value proposition. The tell is that the outcome is defined by the seller. And beneath that, there is a colder force the debate keeps ignoring. Meta's token pricing is not a competitor to per-outcome billing. It is the market-clearing price that makes per-outcome billing impossible to sustain. When a token costs two dollars per million, "cheap and coarse" will beat "precise and expensive" in every procurement meeting where the buyer is honest. The industry is not converging on value-based pricing. It is converging, unwillingly, on the price set by whoever owns the cheapest intelligence. That is the blind spot in every optimistic forecast: the debate is happening as if the participants will choose the winner, when the winner has already been chosen by the cost curve. What I want you to hold onto is not a price. It is a question. If autonomous agents are to buy and sell from one another — if the value-transfer pipes are already running and the settlement is already climbing — then we are about to need, urgently, a verifiable answer to a question law has never had to answer: when a machine claims an outcome, who confirms it, and under whose authority? The curators of this next layer — the ones who build verification with the same care we once reserved for identity — will decide whether the six dollars and seventy-eight cents becomes shared prosperity or simply a smaller number for everyone. Curating the soul in a world of derivative clones was always the harder job. It is about to become the only job that matters.