This actually wrong way to do it, here the pro way:

1.Start writing a game/app/engine/os.

2.Write dozen utility packages for it.

3.Write high-level packages for #2

4.Write a “light-weight” version of #3 when disgusted with bloat of #3.

5.repeat #2 with more modern rust code.

Linux lacks GUI configuration tools for many things, you have to edit text files often using guidance for obsolete versions of software and hope it works. Every single config file can have thousands of lines and if you wrote something wrong it will crash or start acting weirdly, very fragile design. GUI config tools mostly allow valid inputs like checkbox true/false and complain if the path isn’t valid.

Edit: to clarify, i’m exclusively using linux since 2008 and i’m not ‘afraid of editing config files’, downvoting me doesn’t fix the problem. I’m also not fond of fixing your header files for them to compile.

The underlying idea https://en.wikipedia.org/wiki/Vactrain has been tried many times, but its simply too expensive and fragile for real-world use. A single earthquake can cripple it.

AI that designs a room-temp superconductor,quantum CPU,some new chemical that allows cheap/easy solar panel construction, water desalination method/materials,new drugs/compounds,etc. The most important areas are fundamental physics/chemistry/material inventions, if those could be brute-forced by AI search/filtering.

The next area where i think we could replace dangerous jobs with finer robotics: Currently the robot AI is too primitive for fine movement and precise grasping, but imagine automating everything from chemical experiments to nuclear reactor repair with micron-level precision.

A third area: AI could reinvent software testing/debugging, instead of “debugging” some AI module could trace every error and provide clues as to what happening, no more chasing stack traces and adding print to every single line. I can envision 99% of manual debugging drudgery can be automated, even crash analysis. AI doesn’t have to understand code or to write it, just understand the control flow and what is currently executing. AI-generated fuzzing, unit tests and automated cross-platform testing, instead of arcane scripts that break every second release and have to be fixed for each new features, it could be unified into some AI service that recognizes unique requirements and run automatic tests/analysis tailor-made for the target code.

On the creative side, i hope the content generation will stop being so energy-wasteful and slow, i don’t want to buy graphics cards for this but it seems every neural network/image generator even simple upscaling apps and voice generators require a NVIDIA card(AMD seem to have problems).

I have to write userscripts to make websites bearable, software UI is not as customizable( bright white background at night, tiny 9pixel fonts, huge empty margins and content crammed into 200x200 boxes,lots of tiny cryptic icons crammed into toolbars that suppose to be menus,etc )