I’ve decided to move away from the excessive AI bloat taking over. I still use AI however I prefer to have more control of what I agree to use. My plan, since my disability gives me a great use case for AI, has moved towards becoming an AI Minimalist, instead. For personal projects, I’ll still use the online models while keeping up to date on the latest dev stacks through studies to prototype, then swap over to my offline local models for assistance when needed as I clean up the massive technical debt on large projects manually.

Yep, that’s right, I’m tired of excessive dirty code added to my projects.

I chose to move to a more privacy-oriented approach. As an AI Minimalist “less, is more“; strip away unnecessary complexity, maximize efficiency, and finally ensure AI systems stay transparent, controllable, and aligned with human values. I don’t need AI models to have full control of my OS and applications; I can do more selective integration through n8n, if I see fit, through local ollama calls.

Don’t get me wrong, I’ll still use them for work, it’s awesome to implement needed changes with haste on massive projects. I’m just unimpressed by some of the troubles for proper privacy. I’ve found that “Yay, this fix was quick,” multiple times over hours, brings with far too much technical debt into projects when it starts to change too many files at once. I spend way longer code reviewing the LLMs changes than progressing properly. It doesn’t feel right to see blatant backdoors and unsafe coding practices added on the regular, then usually needs extensive hours of work to restructure.

In discussions with friends, and colleagues, it’s also definitely troublesome to have newer developers with less experience throwing hundreds of files changes into single commits with way too little testing.

The data these models are trained on has far too many bad coding practices, is my assumption. It always helps bring rapid development to a feasible proof of concept, at least; however, it also keeps bringing bad technical debt the average dev these days never seems notice.

I’ve also found it’s troublesome to have my entire chat history in the context every time I ask anything, no matter the context of where I talk to the models. Consider how almost always I get asked along the lines of “do you want me to flesh out a n8n workflow, python script, or sql insert, for this?” for any question, around any code idea. I’m usually just trying to make sure I thought things through well. Just because I spent several hours expanding my n8n knowledge while I implemented my own workflows doesn’t mean n8n is what I’m using every time I ask a question that might hint at workflows. Nope, I’m over it.

I enjoy my minimal context when chatting with my local models, I can get any new clean context at the snap of my fingers. It’s great when jumping between tons of projects to get a clean context per feature, I don’t need it to reminisce on a few months ago I was also doing some work on data in a certain style, and having it throw that style in this context where I just want the base structure.

Today (30 Dec 2025), I decided to do my first step, swap my main OS to a Linux distro. I’m keeping my Windows, possibly, will make sure it’s up to date monthly; to keep my access to some software I own and have licenses for. I dislike using too much through wine, it’s great, but with all software getting AI bloat these days, I want my main OS to stay more private. Yeah, it took me 2 days with the limited evening time to iron out what I felt I wanted.

As a note, this isn’t a guide to dual boot your own Linux distro, just to help you consider if you also want to take the steps to move over yourself. There are always more options, I just took the time to consider what I needed and decided what I’ll do.

I decided to go with Pop!_OS by System 76 24.04 LTS with NVIDIA. It’s based on Ubuntu, hence Debian, and open source.

Then for my usual tools of choice, I decided to distance from certain software, and replace others. The end result, where I hope I didn’t miss something I need, is as follows:

• Browser: Edge – Brave would have been the go-to but it wasn’t working well, so I stuck to Edge, for now.
• Gaming: Lutris / Heroic / Steam / Minecraft – A little extra work for installs of launchers, and the likes, but not bad.
• Cloud: OneDrive – I’m still content with all the features and auth.
• Code: VSCode / mono-complete / Android Studio (emulator) – Continue extension to use ollama API AI assistance. Still using C# 12, I’m sad parts of MAUI under Linux aren’t available, but glad I got to work on my Android apps.
• AI: ollama – Local models for the win!
• Services: Podman – My personal n8n, and kanboard, at present.
• Password Security: KeePassXC – I’ve moved to better security these days.
• Video Editing: kdenlive – Appears to have features I used in my previous editor.
• Recording: OBS – This should be obvious.
• Office 365 – Through the web, whenever needed.
• Game Dev: Godot .Net – Working well in preliminary tests.
• Media: Cosmic media player – Needed a few modules installed for encoding, seems decent.
• Notes: Obsidian – Organisable collections of notes, studies, my collections of data, and the likes.
• Chat: discord – In my few communities I enjoy.
• Image Editor: pinta – Feels pretty awesome, just undecided still.
• Screen Clips: Shutter – Only full window at present; still undecided if I like the Cosmic UI of the OS. Cosmic UI crashes a little too much so might move to X11, there appear to be some awesome themes.

I can’t think of other things I might need or use further than these, at the moment. At least, since I can still boot to my Windows 11, for the foreseeable future, there shouldn’t be any problems.

I did try to make a /boot/EFI/Windows11 entries and systemd-boot ignored configs, grub also didn’t work at all, but booting straight through my BIOS boot options worked perfectly. My Photolemur 3 might run under wine, so will take a look at that more in the future.

You can understand, for the lesser ML use, 1.5b models are just to push my imagination through, they’re convenient with speed. I swap between the 7-8b models when doing certain code adjustment queries, and 24b or larger I do massive prototyping with on older style projects. As mentioned above, new tech I try keep up with through the web models, for a while. Though I’m more focused on cleaning up dirty code from models these days, derp.

As a side note, I may have missed something, but I used below commands to get MAUI Android dotnet workload working through VSCode.

winetricks --force -q dotnet40
winetricks --force -q dotnet45
sudo dotnet workload install maui-android

Then to create a MAUI app, the command below, then in the .csproj remove, or comment out, the other platforms. I’ll only target Android on my Linux but can uncomment platforms on Windows 11 and build there, if I feel like it. There were a few things needed to make the environment vars work for Android Studio emulator use, but I got there.

dotnet new maui -n MyMauiApp
cd MyMauiApp
dotnet restore
dotnet build -t:Run -f net10.0-android -p:AndroidSdkDirectory="$(realpath $HOME/Android/Sdk)" -p:JavaSdkDirectory="/usr/lib/jvm/java-17-openjdk-amd64"

With all that shared, I’m happily still doing what I love, as a full-stack dev, and still expanding my knowledge around AI as a whole. I’m rather just not letting it take over my computer so I can keep my own privacy a little more.