Co-authored with Claude. I had scattered thoughts on this topic and we worked through them together.
I’ve been noticing a trend: endpoints built specifically for the frontend are becoming the norm. Tighter coupling between client and server, on purpose.
This shows up from two directions.
From the frontend side: meta-frameworks. Next.js and SvelteKit let you run code on the server before the page loads — getServerSideProps, load functions. You can call your database directly, validate and transform data, all in a trusted environment. The frontend reaches into the backend.
From the backend side: HTMX and the likes. The server returns HTML directly. Links and buttons encode what actions are available — if the server renders a “Delete” button, that is the authorization check. The backend reaches into the frontend.
I tried HTMX because their essays make the case so clearly. It’s novel, and I liked the purity of the argument. I built a small dashboard with Axum, Maud, and Alpine.js for client-side bits. Simple CRUD worked fine. But once I needed a filtering DSL — composing partial updates, managing state across interactions — I got lost. In the end, it didn’t match my brain. The React model (data → UI as pure functions) is still how I think.
If you don’t want that tight coupling — if you want a separate server-client architecture — I think typed RPC makes sense.
At HubSpot, we use an internal framework called CHIRP. It’s Avro-based: define your service schema once, get typed clients for Java, Python, and TypeScript. Frontend and backend share the same contract. The compiler catches mismatches before production does.
tRPC does something similar for TypeScript-only stacks. No code generation — just inference. I’m excited to try it in a hack project. Heard good things.
The real appeal of all these approaches — meta-frameworks, HTMX, typed RPC — isn’t architectural purity. It’s velocity.
When frontend can pull data without waiting on a backend ticket, features ship faster. When backend can render UI without coordinating with a frontend team, same thing. The best setup is whichever one lets your team move without blocking on someone else.