LLMs as Language Compilers: Lessons from Fortran for the Future of Coding

The article pitches a bold vision: today’s large language models (LLMs) aren’t just chatty—they’re turning into robot chimps that can build entire apps and maybe whole teams’ worth of work. The author says coding agents helped whip up a full iOS prototype in an afternoon, but also tells a cringe story where the bot insisted on the wrong setting like a stubborn intern. Cue the comments: half jaw-dropped, half eye-rolled. One skeptic snarked that claims of “team-scale” output are fiction, while others argued this is the dawn of a new era where Stack Overflow is out and AI copilots are in.

Then the real fight broke out over complexity. One camp says accidental complexity—the busywork and glue code—is falling fast, and wonders if AI can tackle the essential hard stuff next. Another fires back that accidental complexity is actually skyrocketing, because cajoling a bot without theory just spawns more mess. A history buff waves the Fortran flag, reminding everyone that when computing gets cheaper, it gets democratized—gatekeepers always get humbled. Meanwhile, the community turned Steven Yegge’s [Gas Town] meme into a running joke: super-smart chimps who can “wreck your stuff” in seconds. Between “robot chimp” riffs and “philosophical zombie” zingers, the thread was equal parts future-shock and roast session.