Per-call pricing for AI skills: why pay-per-use beats buying the file

Today you buy an AI skill the way you buy an ebook: one price, one download, yours forever. Pay-per-use flips that — and once you look at the economics from both sides, it's hard to defend the old model.

What the one-time sale actually costs each side

For the seller, a one-time sale is a one-time payday on an asset that took real work to build — and because the buyer receives the file itself, that single sale can also be the moment the asset escapes into the wild. Best case, you earned $15 once. Worst case, you earned $15 once and your method is now circulating for free.

For the buyer, it's a blind purchase. You can't try the skill before buying the file, so you pay full price to find out whether it's any good. And you pay the same whether you use it once or a thousand times — which means light users subsidize heavy ones.

What per-call pricing changes

In a pay-per-use model, the skill stays on the marketplace's cloud and buyers pay a small amount each time it runs — think cents per call, not dollars per file.

For buyers:

• Trying is cheap. A first call costs cents. If the skill disappoints, you're out pocket change, not the full price of a file.
• Cost tracks value. Use it three times, pay for three calls. Build it into a daily workflow, pay in proportion — and by then it's earned its keep.
• Agents can buy on demand. When your AI agent hits a task it can't do well, it can call a specialist skill mid-task for cents, without you committing to anything.

For sellers:

• Revenue recurs. A skill that's genuinely useful gets called again and again. Illustratively: a skill priced at ten cents a call, used by 50 buyers averaging 30 calls a month, earns more every month than most one-time prompts earn in their lifetime.
• The asset never leaves. Call #10,000 is as protected as call #1, because no call ever delivers the file.
• Price discovers itself. One-time pricing forces you to guess the lifetime value of your skill upfront. Per-call pricing just charges for actual use.

Why pricing has to be model-aware

One honest subtlety: running a skill isn't free for the platform — each call invokes an AI model, and models differ enormously in cost. A fast, light model costs a fraction of a cent per call; a frontier model can cost ten times that or more. A marketplace that charges one flat price for every skill either overcharges simple skills or loses money on sophisticated ones.

So per-call prices should reflect the model tier a skill runs on: lightweight skills priced lower, heavyweight skills priced higher, every tier sustainable. Sellers pick the tier that fits the job; buyers see the price per call before they spend anything.

The model only works if the file never ships

Pay-per-use has been tried with downloadable products, and it always collapses to the same problem: if the buyer holds the file, why would they pay per use? Per-call pricing isn't a billing feature you bolt on — it's the natural pricing of sealed, server-side execution, where there's no file to own and the thing being bought really is each individual run.

That's the model Sealed is built around: your skill runs sealed on our cloud, buyers pay per call, and the file never ships. The remaining question for sellers is which kinds of skills this protects — worth reading before you decide what to list.