The VPRI STEPS project tried to do this in 20,000 lines of code, and to me it seemed like they got close, but no cigar. But things similar to this have been done several times before by small teams and even individuals: Oberon by Wirth and his students, Smalltalk by the Learning Research Group at Xerox PARC, Plan9 at Bell Labs, and, most notoriously, TempleOS by one schizophrenic guy who thought everyone who made him uncomfortable was Satan, so he ended up homeless and then killed himself.
And we have access to much better technology than they do: teraflops supercomputers on our desks, 32-bit RISC chips that run at 50 MIPS and cost 14¢, decades of advances in formal methods and other aspects of computer science.
Although Eleanor is certainly a very knowledgeable person whose opinions merit deep consideration, I don't think it's nearly as technically hard as she does. I wrote Ur-Scheme in three weeks, and that was before AI.
@xentrac your comment is also showing up without context on mastodon.social. Reading the context, Eleanor's quip/estimate of 41 years (for 100 full time developers plus support) was based on the parent comment's constraint of "One line of code per developer day sounds like an appropriate pace to write software".
On Mathstodon, I see that my comment here is presented without the context of Eleanor's original comment: https://infosec.exchange/@dymaxion/117588732908802911