Building a SaaS Boilerplate Solo in 2 Weeks with AI

NestJS is the most popular enterprise Node.js framework, but it had no ShipFast equivalent. Next.js developers have Supastarter, MakerKit, ShipFast — NestJS developers had nothing production-grade. So I built one.

The Gap

I spent two days researching every NestJS boilerplate on the market. Seven paid products, all with the same problem: they were module collections, not integrated systems. Auth didn't talk to billing. Teams didn't talk to admin. You'd buy the starter and spend weeks wiring things together.

Scope Discipline

The biggest risk in a solo build isn't velocity — it's scope drift. I created three guardrails:

1. A product-understanding document that defined what NestForge is and isn't

2. A decision filter: every feature request had to pass a "does this serve a solo developer shipping their first SaaS?" test

3. A claims registry: every marketing statement maps to code evidence

AI as a Team Multiplier

Claude Code wasn't just autocomplete. I used CLAUDE.md and AGENTS.md to give the AI context about the entire system — architecture decisions, naming conventions, module boundaries. The AI could make informed decisions because it understood the product.

What AI is bad at: product judgment and scope discipline. I built a "modes" system — leadership, product, engineering, marketing — to keep the AI focused on the right layer at the right time.

The Numbers

- 227 passing tests across 21 suites

- 0 TypeScript `any` types — strict mode enforced

- 10 integrated modules working together on first checkout

- 38 verified claims in the claims registry

- Built and launched in under 2 weeks

What's Next

NestForge Pro is live at [nestforge.dev](https://nestforge.dev). I'm writing more about the technical decisions behind it — starting with why I didn't use Clerk for auth.