Did you have to look or review any of the code produced, to get the performance/capabilities that you wanted, or were all interactions through CC? In other words, did you hit any walls with the pure agentic workflow?
How does this work? Files need to reference other files eg. for calling functions from other modules, which means semantic analysis needs both files in memory to check the types. This is especially complicated with mutual recursion across modules (separate compilation doesn't apply here). If you're building a language like C where everything requires forward declarations, then maybe, but anything more modern seems difficult.
I think you’re right with regards to the intention — but I’ve personally not experienced the case of an std lib being too big — good examples of “the right size” would be Go or Zig.
The fact that you either need a third party dependency or a large amount of boilerplate just to get decent error reporting, points to an issue in the language or std library design.
I've started also dropping `thiserror` when building libraries, as I don't want upstream users of my libraries to incur this additional dependency, but it's a pain.
For sure, and I guess that's kind of my point -- if the OP says local coding models are now good enough, then it's probably because he's using things that are towards the middle of the distribution.
In my experience the latest models (Opus 4.5, GPT 5.2) Are _just_ starting to keep up with the problems I'm throwing at them, and I really wish they did a better job, so I think we're still 1-2 years away from local models not wasting developer time outside of CRUD web apps.
I do think Linux is good enough (it's what I use daily), but I'm a software person. I don't think linux is good enough for the average person or for kids learning how to use computers. It's hard to use and quite unfriendly, even distros like Ubuntu.
The second point is simply that I think we can do better. Progress does not stop here, but most people are afraid to take on big problems, so they never try.
In five years time, "I want AI" will be 99% of computer users. Sure, neural nets are opaque, but having an AI assistant running locally and helping you with your tasks does not make your computer any harder to understand.
Thanks for the feedback. My goal at this stage is to put the full vision out there, and refine it to create a sense of direction, a north star under which to work. A lot of the specifics are undecided/unknown and that's ok at this stage. I'm building this bottom up using the skills I have (software), and as I bring on hardware folks, those aspects of the system will start to clarify as well.
Whether the project is bigger than I think or not is not so relevant for me personally, I will attempt it because I don't see a future in personal computing that I want to be a part of otherwise.
I agree! I've been thinking about why terminal software is so compelling, and how to make that the default while keeping the system accessible to beginners. I think there's a way to do it, to unify GUI, TUI and CLI.
Thanks for your comment! In terms of concurrent programming in Radiance, it's likely I go for something inspired by Go's simplicity and Haskell's power with STM[0]. Actors are also on the table, but likely as a library on top of the native system, whatever it is. The important thing is that everything that involves "waiting" be composable in this system: timers, network i/o, IPC, file i/o, etc.
For the AI/OS intersection, it is indeed a very interesting design space. The key insight really is that the better the AI knows you, the more helpful it can be to you, so the more you give it access to, the better. However, to be safe, the OS itself needs to be locked down in such a way that personal data cannot leave your device. This is why capabilities-based security is an interesting direction: software should not have access to more than what it needs to operate, and you need fine grained control over that.