The old adjacency of 'SQL for relations, Mongo for flexibility' is dead. Postgres can do it all, and probably faster.
The Convergence
PostgreSQL has evolved into a multi-model database. You need relational integrity? It is the best. You need schematic-less documents? JSONB is faster than MongoDB in many benchmarks. You need vector search for AI? pgvector is the industry standard.
Operational Simplicity
Running a single database simpler than maintaining a polyglot persistence layer (e.g. Postgres + Mongo + Pinecone). Simplicity in operations correlates directly with uptime.
When NoSQL Still Wins
There are niche cases: ultra-high write throughput (Cassandra), massive graph traversals (Neo4j), or specific document-heavy workloads where MongoDB's developer experience (DX) wins. But for the general purpose startup stack? Just use Postgres.