vrypan |--o--| pfp
vrypan |--o--|
@vrypan.eth
Software development frameworks will have to rethink their role (or even what they mean) in an AI-assisted world. In 99% of the cases, the frameworks we use contain code that's irrelevant to our use case, and they are architected in a way that almost never fits our own needs 100%. So, why use a framework, when you can have an agent spit the code you need in minutes, tailored to your specific needs? Maybe frameworks will become instructions for LLMs: Structure it like this, make sure you put guards for these cases, handle these events like this, make this part configurable, do not assume that X will happen, etc. And maybe they also include clarification questions: Should this be able to run in parallel? Do you need event logs? And then we will just ask something like "I want an HTTP API listener, following the <what use to be the package name> guidelines, that will allow me to list/add/remove entries" and get the code that we need exactly.
1 reply
0 recast
11 reactions
Tony D’Addeo pfp
Tony D’Addeo
@deodad
agree and disagree I think it destroys mediocre frameworks; ones that are too thin to add value (i.e. just ergonomic), ones that try to wrap too broad or nebulous of a problem, etc; many things that are better hand rolled are now possible to hand roll without a huge lift but good ones, for example viem to interact with an eth provider with strong types, higher level apis, retry and fallback are just as useful to an LLM one it lets them move faster and two it makes it legible to developers, e.g. I'd rather have the LLM define my server using a simple, well structured, strongly typed web server than hand rolling it all from node primitives
1 reply
0 recast
1 reaction