Recurring, By Agreement

There is a phrase that appears in diplomatic readouts after difficult bilateral meetings. "A frank and candid exchange of views." Read literally, it sounds almost promising. In practice, every embassy in every capital understands exactly what it means: the two sides did not agree, the conversation was not productive, the relationship is strained. The phrase has been hollowed out by a century of use until it carries something close to the opposite of its literal meaning, and the system keeps using it because nothing else does the work it does, which is to communicate bad news without naming it, so the parties can return to their capitals and the meeting can be added to the record without becoming a crisis.

I have been thinking about that lately because the place I keep seeing the same instinct is not in foreign ministries. It is in the way the word "recurring" has come to be used in Tech.

There is a board meeting I keep coming back to, from the months before a raise. The company did agency work, design and integration and account management, and they had built an internal piece of software the team used to run delivery. Margin was good. Growth was steady. The problem was that "agency" was not the word that priced. They needed ARR. What got decided in the room was that the contracts would be split. Part of the bill would stay labeled "managed services." Part would be relabeled "platform subscription," on the theory that the software did some of the work and the client was, in a sense, paying for the platform alongside the team. The MRR line would carry the platform half. The pitch deck would carry the MRR line. The investor model would carry the multiple.

The giveaway came late in the meeting, almost as a side comment, when somebody on the team said: "if the client ever wants to use it themselves, we'll just spin up a tenancy."

I think about that sentence a lot. It told you everything about what the decision actually was. The platform was not being used by the client. It was being used by the agency, for the client. Calling that half of the bill subscription revenue required everyone in the room to agree that the platform existed in a more independent state than it actually did. The strange thing was that everyone did. The founders, the board, the bankers in the next room, the investors who would read the deck. The decision moved. No question was asked. Nobody flagged the seam.

I am not writing about that meeting because it was unusual. I am writing about it because it was not. What happened in that room is not the gentle version of the recurring-revenue confusion that gets written about elsewhere, where a founder uses the word loosely because clients keep coming back and nobody corrects her. That gentler thing is real, and I have seen it. This was something else. This was a room that knew, with full clarity, what kind of revenue sat in front of it, and chose to extend the word anyway, because extending the word was what made the deal possible. Every person in the room could have told you why "platform subscription" did not really apply. Nobody did.

The same pattern is running right now in AI, in a costume that looks new. A company reports $100 million of ARR in eight months. The number is correct. The customer is paying. The math, if you accept the inputs, holds. What is being relabeled is the word "recurring." The revenue is usage, billed per token or per call, with no contractual floor and no annual commitment. Annualizing last month's $8.3 million into $100 million of ARR is the AI version of calling agency work platform subscription. The cash is real. The category is borrowed.

I want to be fair to the founders doing this, because the constraint they are working against is genuinely upstream of them. The capital stack speaks ARR. A Series B investor has a model that takes ARR as an input and produces a valuation. Their LPs expect ARR-based reporting. A founder who tells the room "we don't have ARR, we have reactive token consumption revenue" gets a polite nod and no term sheet. So they translate. They take last month, multiply by twelve, present it as ARR. Each step is rational. Each is what the system asks for. The capital stack is genuinely the constraint, and the constraint is upstream of the founder.

But I keep coming back to a question I am not entirely sure how to answer, which is whether the rooms doing this are unaware of what they are doing, or whether they know and have decided that knowing is not the part of their job that matters.

The case for unawareness is real. The vocabulary of finance has always been imprecise about new revenue shapes. Bookings, billings, MRR, ARR, NRR, gross retention, net retention. The definitions drift. The accounting rules are written for industries that no longer dominate, and the new industries adapt the old language because the new language has not yet been written. There is a generation of founders for whom ARR is simply what you call the revenue if you charge for it more than once. Pressed on the definition, they would not necessarily land on contractual durability. They would land on "money that came in this month and probably will again." That is not strategic. It is a vocabulary that has been used loosely for so long that the looseness feels normal.

The case for awareness is also real, and I find it harder to argue with. The math is undergraduate. The definition is widely understood by anyone who has read a Bessemer memo or a Salesforce filing. The customer contracts have no commitment clauses and are visible to anyone who opens them. The category mismatch is not buried. It is on every page. To not know is harder than to know. And the people in these rooms are, by selection, the people who are best at knowing what is in the room.

$1 million of monthly usage on pay-as-you-go is not $12 million of ARR. Everyone in the room knows that. The question is why we keep agreeing that it is.

The answer, I suspect, is that no participant in the room is positioned to be the person who says it. The founder needs the raise. The investor wants the deal at the multiple their model produces. The banker wants the transaction to close. The investor's LPs want exposure to AI revenue, described in language their committees can underwrite. The auditor signs off because the math is correct and the category question is not in their scope. The downstream analyst will be the one who points out the mismatch, but by then the round will have closed and the analyst will be writing for a different audience.

Every individual incentive runs the same direction. The single person whose incentive would point at honesty is the participant who is not in the room: the LP, six months from now, after the fund has been deployed, asking why the recurring revenue was reported as recurring when it wasn't.

The cost of the stretch shows up at the next layer of decisions. Once a company has internalized that its $100 million is ARR, it starts behaving like a $100 million ARR company. It pushes for annual contracts where the natural shape of the relationship is pay-as-you-go. It introduces seat-based licensing on a product whose unit of consumption is a token. It builds pricing pages and contract templates and sales motions designed for subscription-shaped revenue, on top of a product whose revenue is shaped like consumption. The metric, accepted as a definition, starts giving instructions. The product accommodates. And somewhere downstream, a quarter or two later, the consumption falls or the renewal does not arrive at the assumed shape, and a different room, the one without the founder, starts asking the question the original room agreed not to.

Lately, companies have begun inventing second metrics alongside the official one. Salesforce, last quarter, introduced something called Agentic Work Units alongside its standard Agentforce ARR reporting. They still publish the ARR figure: $800 million, up 169 percent. But the fact that they felt the need to put a second number on the page tells you what the first number is failing to capture. It is a room admitting, in the only way rooms know how, that the available vocabulary has stopped fitting.

I am not sure how this resolves. Words in finance get renegotiated slowly. Sometimes a regulator forces it. Sometimes a scandal forces it. Most often, a new generation of analysts simply stops accepting the older definition, and the metric quietly loses its weight. Until one of those things happens, the rooms keep meeting, the contracts keep getting split, the metrics keep getting extended past the categories they were built for, and the people in the rooms keep doing the rational thing, which is to nod and move on.

I am also not sure what to do with the question I started with, of whether the rooms know. I think both can be true at once. I think they know in some part of themselves, the way the ambassador knows what a frank and candid exchange of views was. And I think the room makes them not see it for as long as the deal needs. Most likely both, in different proportions in different rooms, and the proportion may not actually matter to the outcome. Whether they know or not, the metric gets extended. The deal closes. And the next quarter, somebody else, in a different room, says some version of the same sentence about the same product, and everyone agrees not to flag the seam.