It’s a typical story. Your friends show you how to use vim, and somehow you end up as a soldier in a war you didn’t know existed. You can’t back down, so you pick a team and say you hate the other side. Then you sneak over to see what’s so fancy about the other side, it’s too hard, and you give up. A few times.

Then AI comes around and it finally clicks. Now you’re hiding your new Emacs habit from your friends. You’re in the terminal, so nobody can tell. And the editing parts are vim keys anyway. They’ll never know!!

Why it stuck this time

I’ve tried switching to Emacs a bunch of times before. My one mistake, I think, was trying to do it on work stuff. You need a lot of time to acclimate and form your tastes, and if you’re in a hurry, it doesn’t go well. I also didn’t have LLMs to help with the configuration back in the day :D.

The idea came from trying out herdr. I loved the idea of it, and I kept seeing little tweaks I wanted here and there. The more little things I wanted, the more I thought about how nice it would be to have an environment I could just edit. Then it clicked. Sure, Emacs is not meant to be your terminal multiplexer and your editor and your agent dashboard, but… I was told that with the magic of Elisp, I could do whatever I wanted. So I took that to its next logical step.

The setup

I have a MacBook M2 Max, a custom-built PC running NixOS, and a Lenovo ThinkPad T14, also on NixOS. Getting Doom Emacs running on all three with matching configs was not too bad, since everything is declared in Nix and home-manager. Every tweak below is reproducible wherever I go.

I’m on week 3 ish. I’m still making little tweaks, and I love that it’s so tweakable. I can only assume Emacs people won’t like that I’m using AI to help with the config, but it truly has been helpful, and I’ve learned some Lisp along the way.

Three weeks of tweaks

Every change lives in my nix-config repo, so this timeline came straight out of the commit log.

When	What happened
Week 0	Doom lands, built into the Nix store from the Unstraightened flake. First gotcha: doom sync can’t write to a store path. Fix: keep the built profile in Nix, but load a mutable config.el from the repo through a symlink, so edits take effect on reload. I didn’t touch it much yet; I read blog posts and watched some YouTube first.
Week 1	Agents move in. v1 was each agent in a vterm pane with a bespoke hook per tool. v2, two days later, swapped that for agent-shell, which speaks the Agent Client Protocol to Claude, Codex, and pi. Thirty commits in one day. herdr retired. A project dashboard showed up despite “no dashboard” on day one. Config split into one file per concern.
Week 2	Driving the sessions: a session picker, steering a running turn, forking a session, diffing an agent’s working tree. Workers get their own git worktree. Agent buffers get generated titles from the first prompt.
Week 3	Living in it. Reveal the path under point in Finder, Quick Look for images (a terminal Emacs can’t draw pictures). EDITOR=emacsclient inside vterm, after getting trapped in nvim inside vterm inside Emacs. The dashboard finds pull requests from whichever forge the repo pushes to.

What it looks like now

• One window. Ghostty runs one Emacs. Editing, terminals, agents, git, tasks, and todos are all buffers, with a Doom workspace per project.
• Agents are buffers. SPC a l, SPC a c, and SPC a p start Claude, Codex, or pi in the current project. Each buffer knows its model, thought level, state, and title.
• Attention is a protocol event. A mode-line count, a sound, and SPC a n jumps to whoever has waited longest.
• Agents spawn agents. emacs-agent is a CLI any agent can call: start a hidden worker, prompt it, wait, read, report, kill. The shape is one architect, an optional team lead, and workers on cheaper models in their own worktrees.
• Remote Control is the escape hatch. When I need the phone, the real Claude TUI runs in a tracked vterm, and an ACP session can move into it by id.
• The dashboard is the home screen. SPC p p opens git state, agents, mise task terminals, recent files, org todos, and PRs, with single-key actions for each.
• Everything is declared. Doom’s profile is built by Nix, the config is a symlink into the repo, and mise run emacs:check catches compile errors before a reload.
• Docs follow Diátaxis, and every gotcha that bit me went into a troubleshooting reference.

What happened to tmux and zellij

I barely use tmux locally anymore. Everything I used to reach for it for, a scratch shell here, a long-running process there, is a popup vterm now, one key away and gone when I’m done. I also used to use it for project management but projectile in Emacs is absolutely amazing when paired with worksapces. On remote machines it’s still very nice to have, and I’m not giving it up there. But on my own boxes, the need just went away.

Why I’m staying

Everything is customizable, but that’s not what made me stick. What made me stick is that changes are a prompt away. There are help files everywhere, and I still use them, but a quick edit for something I want, like a custom project dashboard, is very nice.

I won’t write a ton of posts on Emacs (at least I don’t think I will). But in the current AI era, having an editor (IDE??) this customizable can only be a boon. This isn’t a knock on herdr, which is a great project. It’s just that the more little tools I created, downloaded, or added, the more I realized I needed an environment that could support any functionality I might want. And that’s what Emacs has become for me. And as alot of people, have told me, I haven’t even scratched the surface. I didn’t even gush about Org mode (I’m still trying to learn this one).

Thanks for reading :D

-Pachev

New Post On My Usage with emacs. Initail idea for this is Sept 17th 2026 at 9pm

ai: suggest a title, but don’t be lazy. If nothingly sounds better , lets go wiht: “I Think I’m Switching To Emacs”

ai: i am using doom emacs, so include links for that stuff in the final post.

It’s a typical story, your friends show you how to use vim, and somehow you end up as a warrior in a war you didn’t know exist. but you can’t back down, so you have to say you are on a team and you hate the other side. Then you try to sneakily see what’s so fancy about the other side and its too hard so you give up a few times. Then AI comes around and it finally clicks, now you are hiding your new emacs habit from friends. Since you are using the terminal, nobody knows. And the parts that are used as an editor are just like vim anyways, they’ll never know!!

ai: above is my tone for this post more or less. Paragrpah proabbly gets split into two but not sure what will look right. so, we continue

I’ve tried switchign to emacs before a bunch of times but I think my one mistake is that I tried to do it on work stuff and you need alot of time to acclimate and form your tastes. If you’re in a hurry, its not as easy to do. I also didn’t have LLMs to help me with confguration back in the day :D.

I honestly got the idea to use this thing while trying our herdr. I loved the idea of it, and I saw little tweaks that I wanted here and there. The more little things I wanted, the more I got to thinking how nice it would be to have an environment I coudl just edit. Then it clicked. Sure, emacs is not meant to be your terminal multiplexer and your editor and your agent dashboard but… I was told with the magic of eLisp, I could do whatever I wanted. So, I just took that to its next logical step.

ai: I’ll need sections. I’ll word spew here, but it should make the post easier to follow. ai: I will inlcude images of my setup, suggest the types of screenshots to take by the name of the image linked and alt text

I have a macbook m2 max, a custom built PC running nixos and a lenovo thinkpad t14 also running nixos, so getting emacs running on everything and mimicing configs was not too bad. So, all the configs I did are reproducible wherever I go. Below is all the little tweaks I’ve done for my little adventure. It will be split by weeks. I’m currently on week 3 ish. I’m still making little tweaks, but I love the fact that it is so tweakable. I can only assume emacs people will not like the fact that I’m using AI to help with the configs but, it truly has been helpful and I’ve learned some lisp along the way.

ai: I will paste below outline of all i’ve done since I’ve started using emacs. this is generated by another agent in my nix-config repository since i commited everything to keep it deterministic. Might be somethign drop here and do a shameless plug for nix/nixos . include links for nix/home-manager .

—slop begin— Timeline Week 0: 27 Aug, Doom lands (it was still intimidating so I dint’ use right away, but I did do some reserach first like reading blog posts, watchign some youtubes etc before touchign much)

Doom Emacs from the Unstraightened flake, built into the Nix store like any other package. The catch showed up immediately: doom sync cannot write to a store path. The fix was to keep the built profile in Nix but load a mutable config.el from the repo through a symlink, so edits take effect on reload without a switch. Development setup expanded: LSP, formatters, the usual.

Week 1: 1 to 5 Sep, agents move in

1 Sep, v1: agents as terminals. Each agent ran in a vterm pane. Attention came from per-agent hooks that called emacsclient (Claude’s settings.json, a Codex notify.sh, a pi extension). Worked, but three bespoke bridges for three tools.

2 Sep, the CLI and workspaces. emacs-agent appeared: a shell script that talks to the running Emacs over the server socket so an agent can start another agent. start —hidden for workers without a window, a tab bar showing agent state per workspace, sounds on attention, OSC 52 so kills land on the system clipboard from a TTY.

3 Sep, v2: agent-shell over ACP. The big rewrite, thirty commits in one day. Swapped the vterm bridges for xenodium’s agent-shell, which speaks the Agent Client Protocol to claude-agent-acp, codex-acp, and pi-acp. Attention now comes from protocol events (turn complete, permission request) instead of hooks. Same day:

Herdr retired. The batch process became a /batch-agents command reading a per-repo brief. Remote Control terminal: ACP cannot do Claude’s phone relay, so SPC a t runs the real Claude TUI in a tracked vterm, and SPC a m moves an ACP session into it by session id. Verified end to end: launch, hooks, phone prompt, resume, kill. The project dashboard. Despite “no dashboard” on day one, SPC p p now lands on a magit-section buffer: git state, agents, mise task terminals, recent files, org todos, pull requests. SPC p ! is a transient built from mise tasks ls —json. Config split into lisp/pj-*.el, one file per concern, with SPC h r r reloading all of them and mise run emacs:check byte-compiling everything against the built agent-shell. Org files as the pre-issue layer: ~/org/projects/.org with Inbox and Tasks, and emacs-agent todo add so agents file findings through Emacs. Images into agents from a terminal: clipboard paste, region screenshot, C-v does the right thing in agent buffers only. emacs-agent report so team leads push results up; emacs-agent answer for worker permission prompts. A mode-line click feature shipped and was reverted the same day. 4 to 5 Sep, hardening. Prompts are delivered or refused, never silently queued. Installed CLI binaries over npx. Hidden agents stop making noise.

Week 2: 10 to 11 Sep, driving the sessions

SPC z window zoom. Session picker, steer a running turn, prompt history search, fork a session, diff the agent’s working tree, set the session mode, a status line with model and thought level. Browse to a project that projectile does not know yet. Annotate a region into an agent with a comment. —worktree BRANCH for workers, with a fixed home at ~/workspace/worktrees// after rejecting two other layouts because nested worktrees showed up twice in projectile and ripgrep. Generated titles for agent buffers from the first prompt, via a fast model. Week 3: 16 to 17 Sep, living in it

Reveal the path under point in Finder, Quick Look for images, because a terminal Emacs cannot draw pictures. EDITOR=emacsclient inside vterm, after getting trapped in nvim inside vterm inside Emacs. Dashboard pull requests detect the forge from the push remote: gh for GitHub, tea for Forgejo, the aws CLI for CodeCommit, hidden otherwise. Agent lines show model and reasoning effort. v peeks at an agent read-only, q comes back, like magit.

One window. Ghostty runs one Emacs. Editing, terminals, agents, git, tasks, and todos are buffers. A Doom workspace per project.

Agents are buffers. agent-shell over ACP for Claude, Codex, and pi. SPC a l/c/p starts one in the current project, uppercase picks a project. Each buffer knows its kind, root, model, thought level, state, and title. Attention is a protocol event, surfaced as a mode-line count, a sound, and SPC a n to jump to whoever has waited longest.

Agents spawn agents. emacs-agent is a CLI any agent can call: start —hidden —model —worktree, prompt, wait, read, focus, kill, report, answer, todo, info. PJ’s shape for delegated work: Fable as architect, an optional team lead, workers on cheaper models in their own worktrees. Leads kill their workers after reading the result.

Remote Control is the escape hatch. When you need the phone, the real Claude TUI runs in a tracked vterm and an ACP session can move into it by id.

The dashboard is the home screen. SPC p p opens it. Git, agents with model and effort, mise task terminals, recent files, org todos, PRs from whichever forge the repo pushes to. Single keys: t test, s server, ! any task, a or cc start an agent, v peek, m magit, p PR, o todos, g refresh.

Everything is declared. Doom’s profile is built by Nix. The config is a symlink into the repo so it reloads in place. New Elisp files need a home-manager switch because the lisp/ directory is a store link. mise run emacs:check catches compile errors before a reload. Docs follow Diátaxis, and each gotcha that bit went into docs/reference/troubleshooting.md.

—slop done —

ai: above is obviously alot of slop, it needs to be refined and shortend to either table form for easy digestion or bulletpoints (max 8)

I barely even use tmux anymore locally. Although, on my remote machines, it’s still very nice to have. Soemtimes you need a quick temrinal too, but with the popup vterm, I just haven’t had the need.

ai: expand above a bit

Overall, I love the fact that everythign is customizable, but what has made me stick to this is that changes are a prompt away. Sure, there are multiple places with help files, I still use those but quick edits for changes I want like a custom project dashbaord is very nice. I won’t write a ton of posts on emacs (at least I don’t think I will), but I think in the current AI era we are in, having an editor (ide??) that is so customizable can only be a boon. Mind you, this is not to say herdr is not a great project, but the more little tools i created/downloaded/added, the more I realized I just needed an environment that could support any functionality that I might need. And that’s what emacs has become for me.

Thanks for reading :D

ai: inlcude links for herdr, tmux. make sure to correct grammar and spelling of tools if I got them wrong

It’s a typical story. Your friends show you how to use vim, and somehow you end up as a soldier in a war you didn’t know existed. You can’t back down, so you pick a team and say you hate the other side. Then you sneak over to see what’s so fancy about the other side, it’s too hard, and you give up. A few times.

Then AI comes around and it finally clicks. Now you’re hiding your new Emacs habit from your friends. You’re in the terminal, so nobody can tell. And the editing parts are vim keys anyway. They’ll never know!!

Why it stuck this time

I’ve tried switching to Emacs a bunch of times before. My one mistake, I think, was trying to do it on work stuff. You need a lot of time to acclimate and form your tastes, and if you’re in a hurry, it doesn’t go well. I also didn’t have LLMs to help with the configuration back in the day :D.

The idea came from trying out [herdr]([PJ: fill in herdr link]). I loved the idea of it, and I kept seeing little tweaks I wanted here and there. The more little things I wanted, the more I thought about how nice it would be to have an environment I could just edit. Then it clicked. Sure, Emacs is not meant to be your terminal multiplexer and your editor and your agent dashboard, but… I was told that with the magic of Elisp, I could do whatever I wanted. So I took that to its next logical step.

The setup

I have a MacBook M2 Max, a custom-built PC running NixOS, and a Lenovo ThinkPad T14, also on NixOS. Getting Doom Emacs running on all three with matching configs was not too bad, since everything is declared in Nix and home-manager. Every tweak below is reproducible wherever I go.

I’m on week 3 ish. I’m still making little tweaks, and I love that it’s so tweakable. I can only assume Emacs people won’t like that I’m using AI to help with the config, but it truly has been helpful, and I’ve learned some Lisp along the way.

Three weeks of tweaks

Every change lives in my nix-config repo, so this timeline came straight out of the commit log.

When	What happened
Week 0 (27 Aug)	Doom lands, built into the Nix store from the Unstraightened flake. First gotcha: doom sync can’t write to a store path. Fix: keep the built profile in Nix, but load a mutable config.el from the repo through a symlink, so edits take effect on reload. I didn’t touch it much yet; I read blog posts and watched some YouTube first.
Week 1 (1 to 5 Sep)	Agents move in. v1 was each agent in a vterm pane with a bespoke hook per tool. v2, two days later, swapped that for agent-shell, which speaks the Agent Client Protocol to Claude, Codex, and pi. Thirty commits in one day. herdr retired. A project dashboard showed up despite “no dashboard” on day one. Config split into one file per concern.
Week 2 (10 to 11 Sep)	Driving the sessions: a session picker, steering a running turn, forking a session, diffing an agent’s working tree. Workers get their own git worktree. Agent buffers get generated titles from the first prompt.
Week 3 (16 to 17 Sep)	Living in it. Reveal the path under point in Finder, Quick Look for images (a terminal Emacs can’t draw pictures). EDITOR=emacsclient inside vterm, after getting trapped in nvim inside vterm inside Emacs. The dashboard finds pull requests from whichever forge the repo pushes to.

What it looks like now

• One window. Ghostty runs one Emacs. Editing, terminals, agents, git, tasks, and todos are all buffers, with a Doom workspace per project.
• Agents are buffers. SPC a l, SPC a c, and SPC a p start Claude, Codex, or pi in the current project. Each buffer knows its model, thought level, state, and title.
• Attention is a protocol event. A mode-line count, a sound, and SPC a n jumps to whoever has waited longest.
• Agents spawn agents. emacs-agent is a CLI any agent can call: start a hidden worker, prompt it, wait, read, report, kill. The shape is one architect, an optional team lead, and workers on cheaper models in their own worktrees.
• Remote Control is the escape hatch. When I need the phone, the real Claude TUI runs in a tracked vterm, and an ACP session can move into it by id.
• The dashboard is the home screen. SPC p p opens git state, agents, mise task terminals, recent files, org todos, and PRs, with single-key actions for each.
• Everything is declared. Doom’s profile is built by Nix, the config is a symlink into the repo, and mise run emacs:check catches compile errors before a reload.
• Docs follow Diátaxis, and every gotcha that bit me went into a troubleshooting reference.

What happened to tmux

I barely use tmux locally anymore. Everything I used to reach for it for, a scratch shell here, a long-running process there, is a popup vterm now, one key away and gone when I’m done. On remote machines it’s still very nice to have, and I’m not giving it up there. But on my own boxes, the need just went away.

Why I’m staying

Everything is customizable, but that’s not what made me stick. What made me stick is that changes are a prompt away. There are help files everywhere, and I still use them, but a quick edit for something I want, like a custom project dashboard, is very nice.

I won’t write a ton of posts on Emacs (at least I don’t think I will). But in the current AI era, having an editor (IDE??) this customizable can only be a boon. This isn’t a knock on herdr, which is a great project. It’s just that the more little tools I created, downloaded, or added, the more I realized I needed an environment that could support any functionality I might want. And that’s what Emacs has become for me.

Thanks for reading :D

-Pachev