2026-09-07 [home]
I have found myself growing more comfortable with generating programs intended only for my consumption in the age of personalized software.
I do a lot of maths research in my free time, and I grew frustrated with rendering LaTeX in my hard when appearing in the terminal. This is a monumentally stupid thing to tolerate. I extended my TUI program to simply bind a small Go server to use another piece of personalized software for rendering Markdown+LaTex: patel.codes/render. This is now the same library that renders my website. It is a completely bespoke LaTeX parser interlaced with a fork of rsc.io/markdown. A few years ago, this would be a deeply misguided thing to do; however, is that still true today?
We have all been there: You setup $SHELL_OF_CHOICE
and $EDITOR_OF_CHOICE, optionally with your
$PLUGIN_MANAGER_OF_CHOICE for choice $PLUGINS. Then,
after $NON_TRIVIAL_AMOUNT_OF_TIME, something triggers
an upgrade somewhere: You deal with breaking changes,
confusing errors, and a wasted day of trying to stand
up your house of cards again.
Working on Go continues to improve my appreciation for the simple solutions. I often find myself asking: Am I the only user? If so, the simplest thing is often to generate the minimally satisfying program and evolve it with use.
For six years or so earlier in my life, I spent a lot of time painting and occasionally sculpting. Nowadays, I mainly shoot film photography as my creative outlet; however, the feeling of crafting personalized software, beholden to none other than yourself, does invoke the similar sense of creativity and fulfillment I experienced when painting or sculpting.
For years, I would go back-and-forth with my teacher, debating the intersections of maths, programming, and art. I wish that he were still around to school me about what's what, especially in this day and age.
patel.codes/mk.shI never knew or cared about editing my dotfiles. I just wanted something that provided the smallest surface area for me to work productively. Over time, it became obvious to me how primitive my configuration was.
I finally took a step back: Why is my setup not just managed by an agent? It exists for my productivity. I know what I want and when I want it. So, I had an agent completely nuke my previous setup. I spent a few hours using the vanilla setup and materializing the key programs that I needed for my workflows.
It was freeing to remove plugin managers for nvim
and tmux while also removing oh-my-zsh. I was
left with something pure: What I needed was the same
as what I wanted.
This culminated in the following library of bash scripts that is invoked by a bootstrapping harness:
05-tailscale.sh # setup tailscaled, optionally with TS_AUTHKEY
10-zsh.sh # no frameworks, just managed ~/.config/zsh
15-jj.sh # with fsmonitor
20-tmux.sh # with custom fzf-based window tui
30-go.sh # download latest go release, bootstrap gotip
40-go-tools.sh # everything i want `go install ...` in my system
50-neovim.sh # built from stable: simple/bespoke plugins
During tight iteration, I was impressed that the agent immediately reached for testing the setup in Docker. I could not believe that I had never thought about doing that before. After modifying the testing harness, it became trivial to drop in and iterate.
Now, I can open any machine and make it uniformly mine by simply running:
curl -fsSL https://patel.codes/mk.sh | bash
exec zsh
Even better, any machine can indepedently checkout
the dotfiles2 repo, make changes, and push it to
the upstream. All other machines simply need to run
d2sync && exec zsh to pick up the changes.
In the spirit of the linked exe.dev blog, this setup is on my GitHub. However, the beauty here is that you need not copy or read my setup; you can simply sculpt your own by pointing an agent at your dotfiles.