Most early-stage SaaS teams introduce a queue, a cache, and a vector database before they have 100 paying customers. Then they hire an SRE to operate all four.
Postgres can be your queue (via SKIP LOCKED), your cache (for most read patterns), your search index (via pg_trgm and tsvector), and your vector store (via pgvector). One thing to back up, one thing to monitor.
We're not religious about this — when you hit real scale you'll outgrow it. But choosing to outgrow Postgres is a much better problem than choosing to operate five services with three engineers.