Been debugging a client's GraphQL API all week. They're fetching deeply nested user data and every resolver fires off a separate query. Sure, dataloader exists. Sure, batch loading patterns work. But somehow every junior dev on the team invents their own broken version first.

REST API with explicit endpoints? You know exactly what you're getting. One query, clear dependencies. GraphQL's flexibility is also its weakness. Some kid writes a query that should take 10ms and hits the database 500 times because nobody documented that User.posts.author needs batching.

The real kicker: I end up maintaining more database code in resolvers than I ever did writing SQL directly. At least with REST I had a clean separation.