Recent Carmack YouTube interview with him saying the code for AGI will be simple:
To be fair, it will most likely be some python imports, for the most of it, with complex abstractions tied together in relatively simple ways. Just look at most ML notebooks, where "simple" code can easily mean "massive complexity, burning MW of power, distributed across thousands of computers".
I don't understand the distinction. A single person can't really write a code now, if you'll disallow ML libraries. You need a complier, libraries, an OS, etc. The vast majority of us work in the highest levels of abstraction possible. That's where the practicality and productivity is. I don't think he's saying "Anyone will be able to sit down and write it in assembly". I think he's saying, it will make sense when it's written, with the abstractions as they are, with hindsight clarity of "we just had to...".