The loss of HTTP/2 Server Push is pretty upsetting. https://evertpot.com/http-2-push-is-dead/

I wish I pushed harder to get folks using it in their APIs. It fundamentally solves so many issues with people jamming bonkers amounts of information into a single payload, like CSS combination but JSON.

Splitting resources to smaller, more targeted endpoints solves the "HATEOAS is chatty" concerns.

It solves "can't cache X because its tainted by data from Y" concerns.

It lets clients pick the data they want in a standard way, without app-level ?include or ?_embed semantics.

You can still do most of this, but being able to say "oh I bet they're about to ask for a X, send that over" is such a bad-ass thing to be able to do. Get some ML in there to spot common client request patterns. Dope, but nobody bothered.

I've been busy help folks describe their APIs from design, mock, build, up to production, and documentation, trying to make OpenAPI tooling (and OpenAPI itself) good enough to do this easily.

Next I was gonna get y'all using HTTP/2 Server Push, but spent too long on .

GraphQL was awkwardly copying Server Push with @defer +websockets, which isn't a path I wanna follow. I dunno. Some folks say Early Hints might be enough but I'm not entirely sure, the Prefer-Push header from Evert was key to not melting ice caps always pushing everything.

GraphQL/JSON:API/HAL all focus on compound documents, which is a great solution to 10 year old problems, but a 5 year old solution (HTTP/2+Push) has been completely ignored and now folks are giving up. https://apisyouwonthate.com/blog/lets-stop-building-apis-around-a-network-hack/