Developers run into all kinds of trouble when they use an orm without understanding the underlying execution model. Before starting Graphcool I lead the team that scaled a consumer site into the top 1000 visited sites globally. A well tuned SQL server and Hibernate took us a long way :-)

One of the biggest foot guns when working with ORMs is lazy data loading. This is a feature implemented by most advanced ORMs that allow the developer to pretend that the entire data model is in memory without having to actually fetch all data on every request. The reason this is so dangerous is that you end up passing objects around that other developers then misuse to fetch "just a little more data". Over time this can lead to tens or hundreds of roundtrips to the database for a single request.

GraphQL tackles this issue head on by encouraging the developer to specify all data dependencies up front. Client side tools like Apollo and Relay even help you collect data requirements from multiple related components and construct a minimal covering query. This means that there is only one request, and the underlying query engine has all information available.

I'm pretty excited to bring these same concepts to the server with GraphQL Bindings: https://github.com/graphcool/graphql-binding