I put off adopting popular agents for most of 2025 primarily because there was no agent-agnostic path to first-class Emacs integration. That changed with ACP (https://agentclientprotocol.com), thus I started working on agent-shell.

I'm happy with how the integration is shaping up, enabling me to have my cake and eat it too (Emacs + AI agents).

I wrote an agent-shell post recently with the latest changes https://xenodium.com/agent-shell-0-47-1-updates

I'm not sure AI Agents can already beat the average VIM user.

I use VIM and I get some inspiration from LLMs sometimes, but me and other developers always after some time give up and just quickly do it manually.

AI Agents will just full the code with maaaassive swaths of texts it will just make a project unusable.

We even get paid to turn client's AI LLM Claude sketches into real programs.

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 swapped from jetbrains products to lazyvim/nvim and honestly without AI it would have not been possible. The speed I was able to configure it exactly how I wanted and support workflows jetbrains would never support is great.

Escaping the god awful AI implementations in both vscode and jetbrains product is also a refresher. They don't know what the ** is going on, they don't know what users want, but I do and AI being able to extend my IDE in a way that helps me be more productive is very undervalued.

May be splitting hairs, but I don't think it's the terminal-native part that's relevant, so much as that both LLMs and emacs/vim are text oriented in ways which e.g. VSCode isn't. (Or perhaps just the text-oriented nature is a result from initial constraint from being terminal-native).

As the author points out, that Emacs is a highly extensible 'operating system' which makes it relatively easy to bring different tasks together. -- This ought to be a natural parallel to what the agentic tools are trying to do (use MCPs and skills etc. to bring different functionality to the LLM execution environment).

That LLMs can help users extend emacs ought to lower the difficulty curve.

Still. It's silly to wish that Emacs could be the LLM's best friend, rather than demonstrating how it is.

RE: "what if in the future all coding skills are irrelevant". My experience has been that good results from LLMs come from putting good thought into its usage. They're quite far from a magic "push the button and get the result you want" where the skill doesn't matter.

I have been a die hard emacs user for 20 years, and I have a very nice emacs setup (if a little bit idiosyncratic, but all emacs configs are idiosyncratic). However, recently I realized that I read code, but almost never write it. What is more, I spend a lot of time doing it in tmux, over mosh, from my phone. Emacs ergonomy is just not great if all you have is a horrible phone keyboard (and no swiping, because tmux redraws the screen if you swipe). And then I discovered helix. It has all the things I was jealous about vim, BUT it has sane defaults ootb. And truth be told, another thing I use a lot is bat, which is cat with syntax highlighting and an automatic pager.

I'm not convinced that terminal orientedness of AI tooling itself will last. My hypothesis is that it was chosen by developers of the current generation building for developers of the current generation. I hypothesize that there is a future where command lines and terminals don't matter, and hence I feel the focus will shift to, as the author points out, to planning, reviewing and ideation tools none of which demands a command line. In fact I expect an entirely new class of tool to emerge that does these things well that is neither an IDE nor terminal based. I think Claude Code's core will live but it's interface will morph in the coming years to adapt to the builders of the next generation. The analogy I use is my kids and manual transmission cars -- they grow up with EVs and single gear drives with linear torque curves, and will have no nostalgia for a manual transmission, engine noises, or supercharger whines. If you never used a terminal, will you pine for it?

Emacs is a true bicycle for the mind. It sets you free in many ways, always wanting to be the tool that simply allows you to do your job more efficiently and effectively.

LLMs are like living with a depressed person who doesn't see the point in doing anything because it's all just been done before. Who cares if you do a good job? Just recycle some shit that others have done and it'll be passable. This isn't a bicycle for the mind, it's just going to drag you into depression with it.

I'm a long time VS Code boy, but have recently started dabbling in Vim because I'm exhausted with AI.

Every VS Code update these days feels like it's exclusively AI orientated. The Vim approach seems fundamentally opposed to an AI-driven development model, or at least says, like everything else, "sure, you can do that, but set it up yourself."

I find the Vim philosophy appealing coming from a work environment that wants to replace me with API calls to American companies.

> AI makes configuration and extension trivial

I recently said to a co-worker:

"Linux is free if your time has no value" is no longer true in the era of LLMs. Just about any wonky Linux issue can be fixed quickly with even just OpenCode and free models.

> Let’s revisit the doomsday scenario. Say programming is fully automated and nobody writes code anymore. Does Emacs die?

Commercial programming will be fully automated. That will not stop people doing it by hand. For all intents and purposes clothing manufacturing is fully automated but some people still do it themselves.

One example near to my heart is my mother. She collects her dog's shed winter coat in the spring, cards it, spins it and ultimately knits it into a coat she puts back on the dog the next autumn - all by hand. She could just buy a cheap dog coat from Amazon, but she thinks it's funny to see the dog wearing a coat made of its own fur so she bought the equipment and learned the functionally useless skills required to do it.

No matter what level of automation is available, a small number of people will still do things the hard way as a hobby or out of perverseness. We might be living in the matrix in 100 years but I'm certain someone will still be trying to work out how to exit vim in their holopod

My time spent in Emacs has *increased* with the increased use of AI. I have to read, edit, accept and reject all those code changes and Emacs+Magit is pretty much the best interface for that.

> Part of the case for Emacs and Vim has always been that they make you faster at writing and editing code.

That was a side-effect of using software that is deeply customizable, but not the main goal. It comes from being comfortable with your tools to the point of relying on muscle memory to use them. The main benefit to me has always been comfort. My favorite commands are a keystroke away, I know how the tool will react and what it's doing at any moment, I know all of its quirks, etc.

None of that changes in the age of "AI". This new technology won't replace computing for me. My hurdle to adopting "agentic" workflows has nothing to do with my choice of tools, and everything with not trusting the companies and ecosystems around this new tech, especially during this insane hype cycle. I've been slowly adopting parts of the stack I'm comfortable with, because, again, my comfort trumps anything else.

Besides, if anything, a deeply customizable platform like Emacs is ideal for building any type of workflow. There are already great packages for LLM integration, and I don't think Emacs users are missing out on anything. The fact "AI" companies decide to build their moats with custom tooling is not a problem I care about. Apple has been pushing their idea of what computing should look like for decades, and it has never appealed to me. I don't miss something I don't want.

Anyway, confusing article. The author acknowledges all of these points, but also the alternative scenarios. Both cannot be true for the same person. There will always be people who enjoy using Emacs and Vim because of their customizability, not necessarily their features. And there will always be people who just want to use a supported and official product, without any tinkering. Emacs and Vim will outlast this hype cycle just fine.

Thanks for all your work, batsov! I do hope you stick around these communities. :)

Agreed on the config. I just launch a Claude code in my vimrc directory now if I want to change anything about my setup or have any questions about how to do something.

Also, I love running a tmux pane for vim, and then like 4 or 5 more -- a few Claude code instances, one for the dev environment, one for interacting with jj/git or other random commands. So easy to switch between tasks. My main annoyance with my setup is that my Bluetooth trackball times out after a period of inactivity, and when I eventually need to use the pointer again there's a lag while it reconnects...

I'm a muggle when it comes to vim, but I've considered learning it again recently because of AI. I'm building more than I have in years because I love being able to try things out without investing 3 months to get something working before I can really test the idea. And so I am typing A LOT. Less code, but lots of markdown, prompts and config. My hands are hurting, I really wish I had a power tool for typing. Writing is always going to allow us to be more precise than speech, and is a tool for creative thought in its self. I can see how we might be bearish on our expectations around new adoptees, but I think there's pressure to get more out of our editors too. Two of my recent projects have been vscode extensions, because I'm needing more help from my editor, not less: https://www.appsoftware.com/blog/fixing-agent-llm-context-de..., https://www.appsoftware.com/blog/as-notes-turn-vs-code-into-...

How much of this article was AI written? I can sense the AI smell all over the place, also I found it way too long with little substance, but maybe that’s just me.

> Imagine an AI assistant that can read your org-mode agenda, draft email replies in mu4e, help you write commit messages in Magit, and refactor code in your source buffers – all within the same environment, sharing context. No other editor architecture makes this kind of deep, cross-domain integration as natural as Emacs does.

I've been a devout emacs amateur for 10 years, and recently learned nvim/lazy.vim because I was tired of web dev kinda sucking in emacs. Sure, you can get a setup with lint-on-save, inline TS errors, and all the other bells and whistles you get in vscode working on a react-tsx project, but then when you `helm-projectile-switch-project` to a vue project, suddenly you have to get a whole new config set up. And oops, for whatever reason, tab spacing isn't being pulled from the editor config for this directory...

But like the author says, LLMs fixed that. I pointed Warp at my emacs config, said my problems, and said "fix plz," and it did. No more "oops," just "fix plz" whenever I'm editing rust, or svelte, or golang, or whatever else for the first time in emacs.

I'm very excited for the possibility in the portion I quoted. I moved away from mu4e and org mode because managing it all was getting tedious: too much time procrastinating by tweaking org configs. Too many emails not rendered properly in mu4e or fat fingered by me and lost. But in world+llm, that's not really a problem anymore. I haven't migrated back to org mode yet but I did an experiment recently asking claude to set up org similar to my trilium set up, and it did a passable enough job that I was convinced it's possible.

So, now I'm back in emacs, trying out the various LLM tools, doing poorly at getting anything other than copilot to work well, and waiting with patience and desperation for someone to make an LLM completion experience in emacs that has a multiline completion experience at least 50% as good as Cursor's.

Agreed.

We should also consider whether the rise of AI for any type of implementation task will reduce the number of new Emacs and Vim users, thereby limiting and ultimately killing the communities' growth.

I came to Vim, and then Emacs, because I wanted a tool for coding that I could configure exactly how I liked. If AI does my coding for me, my need for a custom editor will decrease. More generally speaking, if AI can do any type of implementation task -- coding, task management, email, etc -- my need for software customized for those things will decrease.

If people don't need custom software, many fewer people may seek out and find Emacs and Vim.

Who will maintain them and evangelize once today's generation cannot?

There's an unofficial Claude Code plugin for Vim. It only does plan mode afaik. I haven't used it, but I have considered it.

That said, I've been using Vim as my editor for the last ten years. The last few companies I worked for had integrations with VSCode, so I use it sometimes so that I could show those off, but nothing about it makes me want to give up my awesome Vim setup with my weird color scheme (Dalek).

It took me a few months to go fast with Vim, but once I did, nothing compares. (Yes, I have tried the Vim plugin for VSCode, though it's been a few years.)

RIP Bram; thanks for giving the world this wonderful and timeless piece of software!

I find Emacs better than ever in the age of AI. Install agent-shell to run Claude Code inside Emacs and give it a skill to run Emacs Lisp using emacsclient. Voila - now you have all the integration you could ever need if you are a little resourceful.

Claude Code can now introspect the entire editor and it can manipulate everything. You can have it build whatever features are missing - it will even debug those features for you, live in the editor.

I recently came back to Emacs from VSCode and ended up just letting Claude build me the features I was missing, mainly related to tabs and project management.

I think Emacs has unique advantages in the age of LLMs, you can just mold it like clay.

In my mind I had the opposite picture than the one the article portrays.

Emacs was lagging behind common IDEs, like IntelliJ and VsCode, cause big companies put thousands of developers to combine many features into one integrated package and everything works together providing a very smooth experience compared to Emacs (and Vim probably).

Now IDEs are useless. I personally haven't felt the need to goto_definition or autocomplete variable names for almost 2 years.

Now programming becomes closer to plain text writing and editing and it levels the playing fields for all editors.

Also Emacs can run Rust plugins, the user is not limited to Elisp. Not very convenient but possible.

In my view, Emacs is getting more powerful in the AI era. The cost of writing Elisp has dropped significantly — I've ended up with 15+ personal Elisp packages as a result. The freedom Emacs offers is still unmatched. For example, I heavily rely on sending automatic response (https://github.com/ywatanabe1989/emacs-claude-code), monitoring repository, and targeted context building from dired.

I mainly use (neo)vim because it has a less distracting interface than any other full blown IDE. Not because of some fancy tooling. And because it is faster, but that might be a negligible reason by now.

I purposely try to keep my extension count as low as possible. It's just too distracting for me personally.

If I really want to use AI tools or something else, then I don't mind opening a full suite, but as of right now, I still spend most of my time in vim and use AI mainly in chat mode.

I'm yet to find a satisfying vim AI integration. I want something that blends into my vim workflow, and does not require me to switch Windows and copy paste or reload my open buffers after AI agents edit my code.

For instance I would love for it to seamlessly melt into a "highlight comments/pseudo code" -> some keybind, then AI would expand those to actual code for instance, or I don't know.. but something not like what we have currently

With the idea of the recent HN article that files are the common media of communication between humans and LLM's, it's still useful to have direct file-skills; to be able to read raw files the LLM outputs, or give it input through files, and able to manipulate the files efficiently. You can use an LLM to prompt to search through the files and all that, but for such basic things it's worth learning it yourself, if you do it often.

I'm not about to put any money down - I lack that degree of confidence in my prognosticating - but I doubt the terminal will ever really vanish, for much the same reason that 20 years of touch screens hasn't really put in a dent in a keyboard and mouse for serious work, and game controllers have barely changed despite multiple attempts at VR and other interfaces, and why the stylus is still going strong after more than 5000 years. Sometimes you just get it right.

A text interface is just really damn good at efficient and precise information delivery and interaction, in a way that takes a lot more work for a GUI to match, and they are composable in a way GUIs simply are not. Most users won't - and currently don't - care about terminals, but I doubt it will ever stop being a standard tool for power users.

I don't doubt we'll see new paradigms emerge, but I think they'll come in the form of higher level abstractions for certain classes of task rather than a replacement for the sort of TUIs and GUIs we have today.

This stuff stands on the foundations it's built off of. It's very hard to argue against the stoic determinism of an `ls` call.

And all the success stories I've seen in people using these tools have had a similar theme: top level might be LLM-y but you rush to get to deterministic straightforward building blocks so that you can have reliability.

That, to me, looks like writing up a bunch of small programs to help establish vocabularies and workflows to avoid just churning and getting lost in the weeds.

I'd be interested in seeing some future form of process orientation, but in the meantime.... shells in general have proven they are decently good at tying stuff together quite well.

`ls dir | grep thingy | process` gonna involve less possibility of annoying drift and churn than "run process on all the files with thingy in their name in directory"

Coining this phrase now: "It's the tokens stupid"

Hooking up to and generating calls across filesystem APIs cost multiple orders of magnitude more than calling `ls`. These tooling ideas are interesting, though. Maybe Kenneth_E._Iverson was right all along?

Talking to another senior dev over drinks tonight, we both worried not about our work but about who might come up never having written a single line of code. Never even opened a terminal. Is looking at the code something you learn in semester 5?

I think computer science education is going to stomp onward, poorly. And we will get that generation. And things like "terminal tooling is going out of style" won't even be said any more. Hacker groups will turn from discussions about new ideas to talking about doing leetcode without AI.

Our art died because we used our art to kill it. We are the last human masters.

That's a funny thing to think about.

A terminal still offers a more composable interface than a GUI. Analog feedback is still a concern for high level pilots. You are confusing power tools with entry-level instruments.

I think AI tooling has actually reminded us how useful the terminal really is. running claude code, running formatters and linters and test suites, managing incremental work with git etc. Are all vital in the modern way we write code. I don't see how you can put a glossy UI on top of that and maintain any fine-grained control. Since we're already giving up a LOT of control by having agents write our code, we need to retain some control and certainty over the quality control and review process of the outputs.

I'm somewhat agreeing with it. CLI tools won't go away, since lines of text as an exchange format between tools is extremely powerful, especially with text based LLMs. However using an interactive TUI application to prompt agents will probably be less relevant in the future. E.g. if I want to build a web frontend, why wouldn't I just prompt and interact directly on the side I'm building?

My current assumption is that the interfaces and workflows that stakeholders and product owners use today to manage software engineering resources are the future interfaces and workflows towards agentic engineering systems.

Guilty as charged! :D Short essays are definitely not my forte. (and this one is definitely longer than I wanted it to be, but in my typical style I kept adding to it after I wrote the initial version, and I'll probably add even more content after I go over all the discussion here)