Comment by noctuid
5 hours ago
I like it infinitely more than anything, but it also still feels like a comparably slow, crufty pos that will probably die because of how anti-AI half the core community is. Regardless of how good magit is, it's slow.
I also can't understand how people like eca and agent-shell so much unless they've never actually used any decent harness. Neither seems to really fit with emacs. Acp is the exact opposite of extensible, and eca isn't really extensible either.
Had the last ten years of emacs development never happened, I suspect emacs would be just as used. So no, I don't think the anti-AI stance would extinguish the compelling repl-as-editor proposition that emacs does so well. I wouldn't put too much stock in the agent-shell business. It's got junior dev written all over it, and they're the most vocal.
> Will probably die
Did anyone ever tell you how badly you suck at prophecies? This unfalsifiable prediction has enormously bad track record: people have called Emacs dead since the 90s, through XEmacs, through every IDE wave since, through Atom, Sublime, nvim, VSCode, Zed, etc.
I'm sure "Anti-AI core community" drinking themselves to oblivion and chugging anti-depressants seeing how many packages for AI integration people keep building¹ and they just can't do anything about it. Bruh, there's literally ongoing threads about built-in LLM support in emacs.devel, the latest one is not even a week old.
Magit being slow is almost entirely git subprocess spawn and output-parsing overhead, which exists in any git porcelain, not something specific to Emacs or elisp². Blaming Emacs for that is blaming the UI for the engine. I use Magit on Mac and it is not slow. It is slower than on Linux, but it ain't unusably slow.
Define "decent harness". First of all you're conflating the layers - ACP is a wire protocol; asking a protocol to be extensible is illiteracy - protocols are meant to be minimal and stable, that's literally their job. eca and agent-shell are ordinary elisp, with hooks, advising, keymaps, same mechanisms every other package uses. If that layer is hookable and advisable, it's extensible in every effing sense; extensibility was never supposed to live in the protocol underneath. Darn, I'm extending eca from inside fucking eca buffer, da fuk you even blabbering about? Whatever, skill issue. Next.
¹ https://news.ycombinator.com/item?id=49634649