Notes on billing & revenue

Three pricing models are competing to define how you should charge for your AI products charge for value: per-token, per-agent, and per-outcome. Here's how real companies use each one, where each one breaks, and what it takes to bill this way without guessing.

Stripe Billing is the default choice for a reason. Here's what actually breaks as usage pricing gets more complex, what a migration really carries over, and how to tell if you've hit the point where switching is worth it.

AI products don't have SaaS economics — compute cost scales with usage, so pricing, metering, wallets and entitlements all have to move together in real time. Here's the complete architecture, piece by piece.

Failed payments silently erode revenue and drive involuntary churn in usage-based SaaS. Kribana's autonomous agent diagnoses declines and acts fast, targeting recovery where it's actually likely to work.

At a few thousand events a second, naive aggregation falls over. Here's the ledger design that doesn't.

Prepaid credits look simple until burn-down, grants, and top-ups collide. A field guide to getting it right.

Tier boundaries matter less than most teams assume. Overage framing matters more.

Blind retry schedules recover less money and burn more goodwill than a diagnosis-first approach.

A monthly invoice run that ate three engineer-days became a number nobody thinks about anymore.