March 18, 2026
Robots wrote it, humans lost it
An x86-64 back end for raven-uxn
AI-coded speed boost for a fantasy computer sparks purist backlash
TLDR: An AI agent drafted an x86-64 version of a toy CPU project, making it roughly twice as fast with minimal human time. Commenters split between hardware envy, claims it should just be auto-compiled, and outrage after the project’s community recommendation was pulled, sparking a purity-versus-pragmatism showdown.
A developer just let an AI sidekick crank out a new speed-boosting engine for Uxn, a tiny make-believe computer used by the artsy Hundred Rabbits crowd. The x86-64 version (that’s code for “PC chips”) runs about twice as fast as his Rust build, cost roughly $29 in AI time, and took only minutes of hands-on work. He ran it on a shiny Oxide Computer virtual machine and swears the blog text is all human — the code got help from a “robot buddy.”
The comments? Pure chaos and comedy. One camp is green with hardware envy: “Jealous of the access to Oxide Computer,” sighed one user, joking about needing “three-phase power.” The nerd-fight of the day: a hot take claiming Uxn “could be trivially jitted” — translation: “why not just auto-compile it on the fly?” Others push back, noting Uxn’s toy-and-ROM roots, muddying that “trivial” claim.
Then the drama bomb: a commenter flagged that his recommendation was yanked from the Varvara resources README — a likely fallout from the AI assist, poking the community’s “handmade vs. machine-made” nerve. So now it’s not just speed vs. purity; it’s DIY romanticism vs. pragmatic robots. Meanwhile, the author jokes he should’ve just run with “dangerously-skip-permissions.” The vibe: impressive results, messy feelings, and plenty of popcorn for this tiny-computer telenovela.
Key Points
- •Keeter added an x86-64 assembly backend to his Uxn interpreter (raven-uxn), achieving roughly 2× speed over the Rust version.
- •About 2000 lines of ARM64 assembly were ported to x86-64, initially drafted by the autonomous agent Claude Code.
- •The new backend passed unit tests and a fuzzer used to check consistency with the Rust and ARM64 implementations.
- •Keeter’s earlier ARM64 assembly interpreter was ~30% faster than his Rust version, which itself was 10–20% faster than the reference.
- •The porting effort cost about $29 on an enterprise plan, took a few hours overall, and involved only 15–20 minutes of hands-on time on an Oxide Computer VM.