Stripe Billing was designed for monthly subscriptions and annual contracts. To do anything in between — credits, overage, mid-cycle plan changes, custom enterprise plans, rollovers — you stitch together five API functions, five brittle webhooks, and a handful of database tables you maintain forever.

AI startups change pricing more often than legacy SaaS companies ship features. A new model lands, the tokens get cheaper, the unit economics shift, the pricing page changes by Friday. Every time, the engineering team rips out product/price IDs, migrates active subscriptions, re-checks the webhook handlers, and prays the tax math still reconciles. The cost isn't the migration — it's the chilling effect on how fast the company can experiment with pricing in the first place.

The founders we spoke with had run this loop themselves before adopting Autumn. Two paused in-house billing builds entirely. One had been on Orb and switched because the integration was simpler and the iteration cycle was faster. One picked Autumn because the founders shipped a fix in twenty minutes on a Friday afternoon.^[11]

UBP has crossed the chasm. 85% of software companies have adopted some form of usage-based pricing. 78% of those did it in the last five years; nearly half in the last two.^[4] The growth is no longer a curiosity — it's the default shape of new SaaS contracts.

AI makes seat pricing structurally wrong. An AI product's marginal cost is a token, not a seat. Customers pay for outcomes (drafts, support tickets, agents launched), workloads vary 10× across users, and infrastructure costs move every time a new model ships. Per-seat pricing assumes a stable answer to "how much does this cost us to serve you?" — AI products don't get to assume that.

Coding agents now choose the stack. Developers ask ChatGPT how to ship something. ChatGPT recommends whatever has the cleanest OSS repo, the most idiomatic SDK, and the highest signal on Stack Overflow. Vercel won deployment that way. Supabase won Postgres that way. Autumn is winning billing that way — Apache-2.0, three SDK functions, shadcn-style React components, and an LLM-friendly docs surface.^{[3] [8] [9]}

And the incumbent just confirmed it. Stripe completed its acquisition of Metronome on January 14, 2026 — a reported ~$1B for the metering engine behind OpenAI, Anthropic, and NVIDIA. Patrick Collison's framing: "the shift toward usage-based models will be a defining feature of the next decade."^{[16] [17]} When the rails company pays ten figures for a usage layer, the debate about whether this category exists is over. The open question is who serves everyone below Metronome's enterprise contract floor.

Custom plans per account. A sales-led enterprise plan, a champion's special, a YC discount tier — all created and versioned in the dashboard without a code change. The same primitive powers grandfathering when pricing evolves: old customers stay on the plan they signed up for; new ones see the new one.

Credits, rollovers, and overage logic. Issue credits as monetary value or arbitrary units. Roll unused credits forward. Set overage prices per feature. Workflows that would be weeks of webhook glue against raw Stripe ship instantly in Autumn's UI.^[3]

React components, TypeScript SDK, shadcn-style. Drop-in pricing tables, paywalls, and upgrade flows. The same components every AI dev tool already uses. The integration story is short enough that a coding agent can finish it.^[1]

The reference call pattern. Customers were sold by quick start. They stayed because issuing credits to a frustrated champion, versioning a pricing plan for a quarterly experiment, or stitching together a custom enterprise plan turned into a five-minute job instead of a sprint.^[11]

The economic argument. Engineering time is the most expensive line item at every AI-native startup. The companies that moved fastest on pricing were the ones that didn't need to call an engineer to do it. The dashboard turns billing operations into a non-engineering function — and the non-engineering function is what survives the next reorg.

The pricing power upside. Every customer in our reference call said the same thing: "we'd pay more." Autumn is materially underpriced today. Expanded tiers and enterprise SKUs are on the roadmap, and the dashboard primitives are the natural place to charge for it.^[11]

Stripe Billing serves 300k+ companies and ~200M active subscriptions today — the rails Autumn sits on, and the largest single addressable customer pool in software billing.^[5] 78% of the Forbes AI 50 already run Stripe.^[5] Inside YC, Autumn is already in production at 30+ companies in the current and recent batches.^[2]

The medium-term ICP is every Seed-to-Series B AI-native SaaS company shipping product weekly. They change pricing constantly, can't afford to maintain a billing team, and their CTO is the buyer. Autumn's wedge product is exactly what a YC technical founder ships at week 4 — and the dashboard is what keeps the same customer at week 104.

The category now has a price marker: Stripe paid a reported ~$1B for Metronome's usage layer alone — roughly 7.8× its total capital raised — for an enterprise book of fewer customers than Autumn will touch in its first two years.^{[16] [17] [20]} The acquirer set the comp; the down-market volume play is still unclaimed.