The difference between a prototype and an MVP is that someone can pay for the MVP and keep using it. That means tenants and permissions, billing, an admin panel, monitoring and a release process — the parts that get skipped in a demo and then rebuilt at twice the cost. This is the stack and the staging we use to avoid that rebuild.
The stack, and why it is boring
| Layer | Default choice | Why |
|---|---|---|
| Application | TypeScript, Next.js, Node | One language front to back; hiring is easy; libraries are mature |
| Database | Postgres (managed) | Boring, durable, row-level security for tenants |
| Auth | Managed auth with SSO options | Never hand-roll passwords; SSO unlocks larger customers |
| Billing | Stripe: plans, trials, invoices, tax | Subscriptions are a solved problem |
| Hosting | Azure or AWS, with CI/CD | Predictable cost, your account, your data |
| AI | Claude / OpenAI behind an interface | Swap models without rewriting features |
| Observability | Logs, traces, error alerts from day one | You cannot fix what you cannot see |
Boring is a compliment. Every item on that list has thousands of developers who can pick it up, which is what protects you from vendor lock-in — including lock-in to us.
Twelve weeks, four stages
- Weeks 1–2 — Foundation. Tenancy, auth, billing, admin, CI/CD and monitoring. Nothing a customer sees yet, but everything they will depend on.
- Weeks 3–6 — Core workflow. The one thing the product does, done properly, with the first design partner using it on real data.
- Weeks 7–9 — The AI feature. One feature where a model genuinely earns its keep — extraction, drafting, search, an assistant — with evaluation cases and a cost ceiling per tenant.
- Weeks 10–12 — Onboarding and launch. Self-serve sign-up, onboarding flow, docs, analytics, and the first paying customers.
What we refuse to postpone
- Tenant isolation — it is a security property, not a feature.
- Billing — pricing decisions are product decisions; they need to be tested early.
- Backups and a tested restore.
- An audit log for anything an admin can do.
- A weekly release, even if small, from week two.
AI features that earn their place
The temptation is to sprinkle AI everywhere. The features that survive contact with customers are narrow and measurable: turn this document into structured fields; draft the reply for a person to send; find the record that matches this description; answer questions from the customer’s own data. Each has an evaluation set, a cost per use and a fallback when the model is unsure. Features without those three things are demos.
After launch: run it, or hand it over
Every build ships with the code in your repository, documentation and a production environment you own. From there you can hire, hand it to your team, or keep us on a monthly service for hosting, support and a roadmap sprint. That choice is yours precisely because the stack is boring. It is how we structure SaaS development engagements, from internal tools that became products to AI-native MVPs.
Have a product in mind, or an internal tool that should become one?
Send a short description of the process, or map it in the Process X-ray. You will hear back within one business day.
Leave a Reply