December 25, 2025
When code goes ahuman
Coding Intelligence Asymptotics
AI will churn out alien code — and humans just write the rules
TLDR: Article says AI could code anything, making giant apps where compile time is the choke point and specs matter most. The crowd is split between excitement over “alien code” and fear that writing perfect instructions beats programming itself — with jokes about coffee-length builds and emoji-coded APL.
The latest think piece from The Fiefdom of Files asks: what if AI can code everything, fast and cheap, with humans out of the loop? The author claims mega-sized apps are coming, with compile time — not typing — as the new bottleneck. Forget today’s comfy tools: with “infinite developer time,” machines will switch stacks, invent new languages, even write straight to binary. Specs swell, math proofs check the code, and the real nightmare becomes telling the machine exactly what we want.
Commenters lit up. User ipnon cheered the “napkin math,” then dropped the bomb: we’re headed for “alien code,” like speedrunning’s “ahuman” strategies where reinforcement learning (trial-and-error AI) finds ugly-but-perfect moves. Hype met panic. Some joked we’ll grab coffee while the planet-sized build compiles; others warned that writing specs is the new job, and it’s harder than coding. Skeptics clapped back: real bugs aren’t tidy treasure hunts, and “no dependencies” sounds cute until you’re re‑making a graphics driver. Meme of the day: “Skynet in APL, documented in emojis.” The only consensus? If the machines code, humans become editors of reality — and that’s a drama all its own.
Key Points
- •Debugging/search for bugs can scale roughly as log(N), implying very large codebases remain manageable when coding is automated.
- •With abundant automated development, it becomes optimal to switch tools, create new ones, or emit optimized binaries directly.
- •Machine-created tools may appear alien to humans; current demanding paradigms (APL, Xilinx FPGA, Agda) illustrate today’s time/UX bottlenecks.
- •Specifications can grow in ambition and length; the core challenge becomes defining the correct constraint optimization (alignment).
- •High-assurance development becomes feasible: zero dependencies, formalized specs, mathematical proofs of conformance, and rigorous red-teaming.