Nice writeup! The list of Emacs packages to use LLMs is the roadmap for getting started with AI Emacs integrations and I also liked his list of AI dangers at the end.
I have used Emacs for over 40 years but I found myself getting sucked into the black hole of VSCode as discussed in the article. When I bought a new MacBook a few months ago, I only installed Emacs, no VSCode, and so far that is working out for me.
Also, after many hours of experimenting with AI packages for Emacs, I have reverted to keeping Emacs mostly a pure human activity, and several times a week have sessions with gemini-cli and Claude Code when I want an AI coding assistant. This is just for me, not general advice, but I like to separate use of AI from my regular coding and writing workflows. I use AI for coding cumulatively about three or four hours a week.
I am reduced to keeping Emacs and VSCode open on the same files/project :-)
Use the LLM from VSCode and switch back to Emacs for regular editing!
Currently trying out various Emacs packages. I got 'claude-code.el' actually working. But there is some flickering and also the interface is a bit wonky.
I will try other packages mentioned in the article.
For the life of me, I can't _edit_ with VSCode. My Emacs/Doom/Evil muscle memory is too ingrained.
Install ECA. Read my comment in the same thread. Once you hook up elisp evaluating MCP to your Emacs - it is game over for anything else. Nothing even comes close to the experience. I've been functioning in ADD mode - Annoyance-Driven-Development. I'd get annoyed by some small issue, and without missing a beat, I'd start typing. Well, sometimes I don't even type anymore - I'd just speak and let Whisper transcribe it and Emacs performs some crazy shit. It's not even funny how crazy it is.
YES! I almost forgot about that. What I love about gptel that I can send a request to an LLM from virtually anywhere - any buffer, even minibuffer. Sometimes I "talk" to AI while typing a message to my colleague. I can ask some model to improve my prompt while typing the same very prompt for a different model - that's just bonkers.
Reading comments I feel you guys (Emacs users) have NO idea. I made a tiny MCP in babashka (I could've done it in Go, Rust or literally any other language, but chose Clojure).
So what the MCP does it just evals Elisp in my active Emacs session via emacsclient.
Now, I can tell LLM to change practically any behavior of my editor, and because it's in the REPL-driven loop, it opens up some mind-blowing opportunities. The other day I developed a tiny package on the go, while at some point I asked Claude to run the profiler, find the bottlenecks and fix them - the whole thing was a jaw-dropping complete and utter bananas.
They say: "Well Cursor can understand your code contextually"... Dude! My LLM-infused editor understands not only some arbitrary code, it can touch and improve it's own code at the point where it's touching your arbitrary code. On the fucking go!
Using Emacs was fun before. Now it's gotten crazy awesome.
I have used Emacs for over 40 years but I found myself getting sucked into the black hole of VSCode as discussed in the article. When I bought a new MacBook a few months ago, I only installed Emacs, no VSCode, and so far that is working out for me.
Also, after many hours of experimenting with AI packages for Emacs, I have reverted to keeping Emacs mostly a pure human activity, and several times a week have sessions with gemini-cli and Claude Code when I want an AI coding assistant. This is just for me, not general advice, but I like to separate use of AI from my regular coding and writing workflows. I use AI for coding cumulatively about three or four hours a week.