I have paid for a lot of software I did not use. Everyone has. The forgotten subscription is so normal that whole businesses are built on the gap between what people pay for and what they actually touch. When I priced Gravity, I did not want to build on that gap. So agents are pay-per-use: you buy credits, where one dollar equals a thousand credits, and you spend them on runs you actually choose to make. Here is the full reasoning, including the parts where the subscription model genuinely wins.

It is the pricing-decision companion to how Gravity's pricing works and the broader survey of agent cost models.

The problem with idle seats

A subscription is a bet by the seller that you will pay more than you use, and over a large base of customers that bet pays off precisely because people forget, overestimate their needs, or cannot be bothered to cancel. It is a perfectly legal model and a slightly adversarial one: the seller's best month is the one where you paid and never logged in. I have been on the paying side of that arrangement too many times to want to put customers on it. Pay-per-use inverts the relationship. If you do not run an agent, you do not pay, and my business only grows when you actually get value. That alignment was non-negotiable for me, and it traces back to the same instinct behind bootstrapped economics: make money by being useful, not by being sticky.

Agent usage is bursty

Beyond the philosophy, there is a practical fit. Agent work does not arrive in a smooth monthly stream. You might run an agent forty times during a launch week and not at all the week after. You might use one heavily for a quarterly report and ignore it the rest of the time. This burstiness is the single worst match for a flat monthly fee, which assumes steady consumption that agent usage simply does not have. A subscription forces the bursty user to either overpay in quiet months or feel pressure to manufacture usage to justify the cost. Credits absorb the spikes naturally: you spend when the work is there and you spend nothing when it is not.

Pricing that matches value

The cleanest way to say it is that credits make the price track the value. Each run costs an amount of credits proportional to the work the agent does, so a small task is cheap and a large one costs more, and either way you are paying for an outcome you chose. There is no seat fee, no monthly minimum, and no penalty for irregular use. This is the same usage-based direction the broader software industry has been moving toward, because buyers increasingly want to pay for consumption rather than access (OpenView, 2024). For agents specifically, where the unit of value is a completed task, usage-based pricing is not just a trend, it is the natural unit of the product. You can see how this compares against per-seat and per-token approaches in the total cost of ownership model.

What it does for builders

The pricing model is not only a user-facing choice; it shapes the whole marketplace. Because builders earn a share of every run rather than a cut of a subscription, their income is tied directly to whether their agent is good enough to keep being used. A subscription would let a mediocre agent keep earning off users who forgot to cancel. Pay-per-use does not. If a builder's agent stops delivering, people stop running it, and the builder feels it immediately. That is exactly the incentive a quality marketplace needs, and it is part of why the three-sided model holds together: every side is paid on real value, not on inertia. The builder's-eye view of the math is in the unit economics for builders.

The honest trade-offs

Pay-per-use is not free of downsides, and I would rather name them than pretend. The real one is predictability. A subscription gives a buyer a fixed, plannable line item, while a usage-based bill moves with activity, and some buyers, especially in larger organizations with strict budgets, genuinely prefer the certainty of a flat fee even if it costs more on average. There is also a narrow band of extremely heavy, constant users for whom a flat rate would be cheaper at the margin than per-run pricing. I do not dismiss either. Gravity answers the predictability concern with spending limits and transparent per-run costs, so usage stays visible and bounded rather than surprising. And for genuine heavy-usage or enterprise cases, a flat option could one day sit on top of credits where it actually serves the customer better. The default stays pay-per-use because it is the fairer model for most people, and fairness, not lock-in, is what I want this business to compound on. That is the same principle as letting a user describe an outcome and pay for the result: charge for the value delivered, nothing more.

FAQ

Why does Gravity use credits instead of a subscription?
Because agent usage is bursty and a subscription charges you whether you use it or not. Credits mean you pay for runs you actually get value from, which aligns price with value and removes the dead weight of an idle seat.
How does Gravity's credit pricing work?
You buy credits, where one US dollar equals one thousand credits, and each run costs credits based on the work it does. There is no monthly minimum and no seat fee. You top up when you need to and spend only on chosen runs.
Is pay-per-use cheaper than a subscription?
For most users with uneven usage, yes, because you never pay for idle months. For a small set of extremely heavy, constant users a flat fee can be cheaper at the margin. Usage-based pricing is fairer for the typical case.
What is the downside of usage-based pricing?
Predictability. A subscription gives a fixed line item; usage-based pricing varies with activity, which some buyers dislike for budgeting. Gravity offsets this with spending limits and transparent per-run costs.
How do credits align builder incentives?
Builders earn a share of every run, so they are paid for value delivered, not for a subscription a user forgot to cancel. Income tracks whether the agent is good enough to keep being run.
Will Gravity ever offer subscriptions?
Possibly for specific heavy-usage or enterprise cases where a predictable flat fee serves the customer better. The default stays pay-per-use because it is fairer for most people. Any subscription would be an option on top, not a replacement.

Sources