> Our infrastructure makes it possible to... create custom runtimes for different applications [like] Cloudflare Worker-style Serverless Functions
Fascinated to see what happens here. The serverless / edge compute paradigm fits Javascript hand-in-glove philosophically, but until now it's always felt quite clunky to me. When I've tried it out, I've always been left thinking "but this would just be so much easier with a server".
Reading this has made it click for me why that is. A new paradigm needs a new set of tools native to that paradigm.
The entire server-side JS ecosystem is currently structured around Node, a fundamentally stateful-server paradigm. You can try to abstract over it, but only so far. It's not the serverless paradigm that's clunky, per se, it's that the tools right now were built for another way of doing things.
As a concrete example - Deno has an import system based on URLs, rather than on-disk node_modules. I thought that was a cool feature for convenience, less overhead, more open and sharable packages, etc. But now I realise the full intent of it. It's much more than all that, it's a fundamental shift in the paradigm: no implied dependency on stateful disk in the runtime itself.