@elonmusk@cradrrat I m not a low level programming engineer, but from my current mental model, there is no real benefits to start a greenfield project in C in 2026
@elonmusk@cradrrat I know it could seems like the current vibe coding hype, but what made you choose C over Rust which provides better typing / deterministic feedback loop for the agents, safer memory handling, … with comparable perfs ?
@devinjameson@samgoodwin89 Genuine question, but is that a good timing with @tim_smart currently basically porting foldkit into effect (effect/ui)? I love effect, but I m not quite sure about the current move to absorb every existing good lib built on top. What s the next victim, alchemy ?
@NoemiTitarenco@augustorademake You can prevent most of the issues you listed with types, that is actually exactly why #effect-ts was created.
Your agents need deterministic feedback loop, typing is more important now than ever
@mattpocockuk@EffectTS_ Why not front end code as well ? Atom and refs are a brilliant way to isolate business logic from UI in a react app. Your domain doesn’t depend at all on react at all anymore.
Looking at the Apple keynote, I dont get the 1B more powerful neural engines, heat pipes,.. whereas on the software side there isn't anything revolutionary with Siri. All this machinery to detect ingredients on a photo that we could just do in chatGPT.
How is that now that Github is backed by one of the biggest company out there we have so many outages ?
I mean we can all have our ethics/political opinions on this, but the main advantage of it being backed by Microsoft would be resilience, no ?
@unclebobmartin Your codebase is a golden mine of patterns, instead of throwing agents to watch other agents, I'd recommend combine with more deterministic tools like argot which learns your codebase AST patterns and help you guardrail ai https://t.co/30pHn0YHKs
I built Argot after realizing I was leaving the same review comments over and over:
“we already have a helper for this”
“why this library?”
“this doesn’t belong here”
The code was usually fine. It just didn’t fit the repo.
https://t.co/TAzYIdJkkA
@petergyang Personally I dont think ai can solve itself ai slop, also best code is not always the code of the agent that scored 90+ % at swe benchmark but the one that actually match the codebase / team convention. Tools like argot (https://t.co/26cQFpaUin) are more reliable than skills
@unclebobmartin@sibaiisaac I do think non-ai tooling like argot (https://t.co/26cQFpaUin) to ensure conventions and keep control over the codebase are more relevant than ever.
@unclebobmartin@theo@ori_pomerantz I am 100% with you, but the issue with agents getting smarter and smarter is that they get smarter at cheating too.
I tried to solve this issue but not with another llm skill (agents watching agents… meh) : with the git history and a simple rust binary: https://t.co/26cQFpaUin