I'm curious where backend folks draw the line between "best tool for each job" and just too much operational surface area.

A typical AI app can easily end up with Postgres/document DB + vector DB + auth + object storage + search + LLM gateway + OCR/document service + workers. Each component is reasonable by itself. Together, a small product suddenly has a platform team's architecture.

That pain is why I started Singlebase - one backend API for the common app + AI primitives. I'm not arguing everything should literally be one database/process. More that the developer-facing surface can be unified while implementations stay specialized underneath.

Would you rather have one opinionated backend abstraction, or keep each system explicit?