The biggest problem with depending on Postgres is it's a large complex monolith. Any problem you have only has two possible solutions: 1) spend a ton of time trying to twist yourself into a more pretzely shape to get it to do what you want, or 2) replace that thing you wanted with some external thing.
Both of these waste valuable time on something that isn't providing any business value. They also are entirely preventable/avoidable, by simply not putting all your eggs into one basket.
Basically the whole "just use Postgres" philosophy boils down to: I don't want to learn new things, and I just like making things from scratch with custom code. That's great if you're an engineer and you want job security. But it's bad strategy, engineering, and use of time and resources. Anyone who approves of using Postgres in this way should not be in charge of engineering decisions.