Though Postgres is great, I personally disagree. If you're starting a brand new project now I think you should use something redundant and distributed like CockroachDB or Vitess.
Concretely speaking most startups should be using RDS or whatever hosted Postgres their cloud provider offers, not running their own.
I think it’s sensible to use what one knows best, but for default advice to others I think ”simplest viable option” is generally better.
So does a distributed database. A primary/replica has an ops burden too, and measuring that complexity is subjective.
With primary/replica you still take downtime during failover, you need to make sure you're testing the failover, upgrades aren't zero downtime in a lot of cases, etc.
You're essentially treating your database as a pet and a clustered option is much more cattle vs pet. This saves you time to develop other features (including working on reliability -- the most important feature)