Kristen Maevyn's essay "The Decoherence Trap" names the cost a system accrues when it holds several valid but mutually incompatible models, so every change becomes a job of translating between them and guarding against side effects.
Zig keeps improving as a programming language. Unfortunately, its management increasingly seems determined to make it a bad choice for the people who actually want to use it.
Mike Jackson wants the blind men and their elephant retired as a lesson in systems thinking, and his reasons are good. The parable's flaw is its narrator, who stands outside the world he describes and always has.
Pele, Obsidian's new Mermaid graph renderer, has no DOM dependency and no runtime dependencies, which means it runs on GraalJS on a stock JDK, as well as being a very nice drop-in replacement for client-side Mermaid.
Emergent design says architecture should flow from how something is used rather than from how it was dictated. Inside a codebase, that’s tests pulling the structure into shape. Across codebases it’s something older: a narrow substrate, a caller nobody expected, and a universe that stops exactly where the primitives do.
In "The Geometric Model of Maintenance," someone finally checked the “one month of code, ten days of maintenance” rule against real git history. The number holds up. What it’s made of is worth a look, and so is what you’d need to measure before it tells you anything about your own code.
Nate Meyvis takes up a question AI skeptics keep raising: if AI-assisted coding is so powerful, where is the wave of new, good software? Several of his answers hold up well, particularly that approval chains don’t speed up and that much new software is private. What he doesn’t address is the question itself.
Terence Eden paid strangers to follow his README and found a dozen things wrong with it. Most of them a disciplined authorial pass would have caught first. A few, nothing but a reader catches.
Psychologically, almost every listener decides what you mean before you finish saying it, and decades of writing advice can’t fix that, because that's an aspect of the listener and the advice is aimed at the speaker. A language model is the first listener that takes instructions... and that gives us a way to fix the way we interact with the models.