The tradeoffs with GraphQL aren't worth it

Strong disagree. GraphQL has given our dev teams huge leverage. It's similar to writing a page of HTML rather than the equivalent GUI code. That might not look helpful if you're already a GUI master, but it's a big productivity win for a mixed team of designers and coders. A vague query specification that used to take a week to turn around has now become a precise query specification that we can turn around in under 250ms.

What I like about GraphQL is how quickly one can start fetching data without considering the API endpoint. Especially for nested data structures where the alternative is using json_build_object queries with PostgreSQL. I use it extensively in admin backends and only switch to a rest endpoint when I have to get too creative with custom functions (PostGraphile). PostGraphile doesn't require much configuration, it introspecting your db and you're good to go. The moment the tradeoffs start to show you add a custom rest endpoint.

It's like everything else in the JS world. A new solution to a solved problem. It takes more than an incremental improvement to outweigh the learning curve and development costs. New things should be an order of magnitude better. If a new technology isn't obviously better then it should be ignored. Developers have shiny object syndrome.