Why I built another SaaS foundation when AI can generate one in a few hours

There are already plenty of SaaS boilerplates, and coding agents can generate a surprising amount of starter code now.

So why build another one?

That’s actually the question I’ve been trying to answer while building The Fabrica.

I don’t think the valuable part of a SaaS foundation is “here are auth, billing and a dashboard” anymore. AI is making that part cheaper very quickly.

The part I kept spending time on was everything after the first generation pass:

• deciding how billing state should actually behave

• making webhook processing safe under retries and concurrency

• building metered credits around a ledger rather than a mutable number

• long-running workers and failure handling

• migrations and schema consistency

• account deletion and data lifecycle

• keeping product-specific code separate enough that the foundation can still evolve

The Fabrica came out of infrastructure I had already built for a real SaaS, then extracted into the codebase I wanted to start my next product from.

It’s intentionally opinionated around:

FastAPI + Next.js + Postgres + Celery/Redis + Paddle

and especially suited to products with AI, async work and metered usage.

I’ve also tried to design it for the way I work now: with coding agents. The repository includes extensive architecture documentation, explicit invariants and trade-offs, an AI-oriented guide, and 49 practical recipes so an agent has more than just source code to reason from.

The question I’m still validating is pretty simple:

Will technical founders pay to avoid rebuilding and re-validating this layer, or are the people who understand the value also the people most likely to build it themselves?

I’ve tried to make the trust side inspectable too: there’s a public inspection repo and an 81-second walkthrough recorded from the exact release tag.

Launching tomorrow. Curious what people think on the thesis.