My Coding Setup

I use the following text editors for coding:

• Visual Studio 2022 (yes, that one specifically) for C# app development, since it's effectively the only option for building .NET Framework (up to 4.8.x) apps. Note that the newer .NET apps (>= v5) can be built on other platforms using the .NET SDK, however you don't have access to Windows-specific libraries, such as Windows Forms or WPF, needed for making GUI apps.
• VSCodium for other languages, except when what I'm doing doesn't warrant running such a heavy app (VSCodium is a glorified web browser – no scratch that, it is a web browser – pretending to be a code editor).
• Notepad++ for more basic languages/tasks (e.g. config editing, shell scripts), that don't really need a full-on code editor like VSCodium. I also developed this website entirely using Notepad++ up until May of 2026, when I switched to VSCodium. Note that this app is Windows-only, meaning you need Wine if you want to use it on Linux.
• Nano for all my CLI text editing needs (e.g. configs on remote computers via SSH).
  • Vi, Vim, Neovim, and all relatives have been disqualified because of Vi costing me at least half an hour in total from trying to figure out how to close the damn program, then consequentially figuring out how to change the default editor to something else and having it persist in sudo sessions. I might learn how to use one of these editors one day, but I'm fine where I'm at for now.

Did you know that VSCodium can be accessed from a web browser? To do so, you just need to run codium serve-web from a terminal (or code serve-web if you must use VSCode), then open the URL it shows you in your web browser. Microsoft even provides this as a web service (in a restricted form – for example, no integrated terminal), which you can access here, if you don't mind the opt-out telemetry, for example.

I recently found out about Zed (shortly after its v1.0 release, funnily enough), and am considering using it more instead of VSCodium. The website didn't lie; it's heckin' fast, especially when comparing against VSCodium on older hardware. A lot of features that normally need extensions on VSCodium are also built-in to Zed. I'm pretty much waiting on a couple of annoying quirks to get fixed, or at least get config options (e.g. automatic opening of title bar menus when set to always show on the title bar).

---

I also use the following AI tools for programming:

In case it wasn't obvious from the massive blank space above, I don't actually use AI for coding. This includes both the "big" cloud providers (ChatGPT, Copilot...), and also local AI models. The only exception to this was when I was playing around with local AI models around May 2026, where I found that while it can be useful, it wasn't useful enough for me to justify its continued use. especially given the reasons below.

Note: I do use other tools to assist in code development - for example, VS(Codium)'s built-in IntelliSense code completion. This is not to be confused with "IntelliCode", which actually is local AI code completion built-in to Visual Studio 2022, with a much more limited scope than current-generation AI models, like Copilot.

If you want my stance on using AI LLMs for coding:

• If you don't need AI for coding... just don't use it. If you want some reasons to reject LLMs, see this Codeberg repository, which also lists some projects that allow or encourage AI usage if you care about that. Personally, I avoid AI LLMs for the following reasons:
  • Privacy concerns (most cloud providers train their LLMs on your inputs)
  • Licensing issues with generated code (AIs don't respect the licenses of the data they were trained on)
  • Environmental impact (seriously, some places literally don't get running water any more thanks to AI datacenters being built near to them)
  • AI's role in the job crisis and mass surveillance.
• If you must use AI, run a local model if your hardware permits. To do this, you can set up Ollama, and immediately begin chatting with whatever models are available for download (and can fit in your GPU's RAM). If you also need integration with VSCode, use the Continue extension (also available on OpenVSX for e.g. VSCodium). The built-in Chat feature, while it does supposedly support hooking up local AI models, won't work without signing in to an account, so that's off the table.

If you want a reference for what sort of models you can run, my laptop with a whole 4 GB of VRAM (a number not even mentioned on the websites I used when searching for local AI tutorials) and 16 GB of system RAM was able to run ~1.5-3B-parameter models for autocomplete just fine, and with CPU offloading, could run up to ~8B-parameter models before running out of combined system RAM and VRAM (for more advanced questions/refactors).

Again, and I will repeat what I said earlier, AI will never be used for any of my publicly-published projects, unless I find a truly compelling reason to use it that overrides all the reasons I avoid it above (or I happen to upload one of the test projects that AI did play a part in - unlikely, but you never know). If, for some reason, I decide to publish a project that was written with the help of AI, a disclaimer will be present in the project's README, documenting exactly how AI was used to assist in the project's development.