The AI SaaS launch checklist: 11 things before your first paying user

Most AI SaaS side projects die of the same cause: the demo works, the product doesn't. The gap between "I have a streaming chat route" and "a stranger can give me $19/month" is exactly these eleven items.

Trust layer

1. Real auth with password reset — credentials plus at least one OAuth provider. A login page without a "forgot password" flow reads as a weekend project.
2. A privacy page that names your AI vendor — users increasingly ask where their prompts go. Answer before they email.
3. Status/limits visible in-app — a usage counter converts better than an invoice surprise.

Money layer

4. Webhook-driven subscription state — never trust the client. Checkout session completes → webhook flips DB status. Canceled → same path back.
5. Limits enforced server-side — count tokens per user per day in your own table; check before calling the model, not after.
6. Graceful limit UX — a 429 with upgrade copy beats a silent failure. The paywall moment is a feature touchpoint.
7. Tax handled by a Merchant-of-Record (LemonSqueezy/Paddle) unless you enjoy VAT registration as a hobby.

Ops layer

8. One-command deploy — if deployment needs a wiki, you'll delay every fix. Target: push to main → live.
9. Structured error pages for missing env config — your future self and early users both hit this; a guided setup screen turns a 500 into a 200.
10. Streaming responses with real token accounting — estimate tokens from streamed deltas if the API won't hand you usage on stream; log every call.
11. A kill switch per feature flag — when the model vendor has a bad day, you degrade instead of burning money and trust.

None of this is glamorous. All of it is why buyers ask "is this production-ready?" — and the honest answer is that production-ready is just this list, done.