Co-locating GraphQL queries with the thing that wants the data is truly such an amazing pattern for solving data over-fetching.

The one downside to this approach (that also sort of exists with the old, conventional approach) is that you (the server maintainer) don’t own the webhooks and their queries, so you need some sort of versioning scheme or some way of handling backwards compatibility when you need to break a query. Or you need to never break queries (like Facebook and their mobile apps).