I've seen engineers get genuinely angry about database choices. It's a religion for some people. Here's the pragmatic version: both are excellent databases and the choice usually comes down to data shape, not ideology.
Start with your data shape
PostgreSQL is a relational database. If your data is naturally relational — users have orders, orders have line items, line items have products — Postgres is the obvious choice. The relational model gives you referential integrity, joins, transactions, and 50 years of battle-tested reliability.
MongoDB stores documents. If your data is naturally document-shaped — each entity is a self-contained JSON blob with variable structure — MongoDB's query model fits better. Think content management, event logs, product catalogs with variable attributes.
The JSONB argument
Postgres has native JSONB support that covers most "I need flexible schema" use cases. Before reaching for MongoDB, ask whether Postgres with a JSONB column would solve the problem. It usually does, and you keep the benefits of a relational database for the rest of your data.
Transactions
If your application requires multi-document ACID transactions, Postgres wins on simplicity. MongoDB supports multi-document transactions since version 4.0, but the mental model is more complex and the performance overhead is real.
Scaling
Both scale. Postgres scales vertically (read replicas, connection pooling with PgBouncer) extremely well and handles most workloads without sharding. MongoDB's horizontal scaling story is better for truly massive document workloads. But you'll hit Postgres's limits at a scale most companies never reach.
Our default
PostgreSQL for almost everything. MongoDB when the document model is genuinely a better fit for the core data — which is less common than MongoDB's marketing would suggest.