April 4, 2026
AI's not that into Lisp
Writing Lisp Is AI Resistant and I'm Sad
AI keeps skipping Lisp, devs spiral - from panic to 'just prompt better'
TLDR: A developer found AI breezes through Python but struggles with Lisp, pushing him toward rewriting in Go. The crowd is split between “prompt harder,” “invent better models,” and “we’re headed for a boring one-language future,” with language diversity fears fueling the drama.
A DevOps engineer tried to make AI write his beloved Lisp and got burned: endless fussing with the live coding console (a REPL), ballooning costs, and code he had to rewrite anyway. Then he built a helper tool in Python and—boom—AI flew through it in a day. Now he’s eyeing Go, sighing that AI takes the “path of least resistance,” meaning popular languages win while niche ones suffer. Cue the community fireworks.
One camp says the sky isn’t falling: “Not my experience,” reports a veteran who gets solid Common Lisp results by feeding the bot giant, highly detailed prompts. The tinkerer crew wants new tech: a “graph-capable” model that edits code by understanding its structure (think: a tool that sees the blueprint, not just the words). The philosophers dove in too: maybe AIs, like humans, are wired for certain patterns—so no wonder s-expressions feel alien. And then came the doomer chorus: if AI favors Python and friends, language diversity could flatline. Meanwhile, the jokes wrote themselves—“AI wants chicken nuggets (Python), not veggies (Lisp),” one quipped, as others nitpicked whether Clojure is “lispy enough.” Either way, the vibe is clear: Python parties, Lisp mourns, and everyone’s arguing about the guest list.
Key Points
- •The author attempted to build an RSS reader format conversion tool in Lisp with AI assistance but found REPL-based workflows difficult and costly.
- •They initially used tmux to interact with the REPL, which proved brittle and token-intensive for AI agents.
- •To improve this, they created a Python MCP server (tmux-repl-mcp) enabling direct command execution in the REPL, integrated via uvx.
- •AI models produced significantly better results in Python: full code and tests were generated quickly with cheaper models, unlike in Lisp.
- •Persistent issues include AI defaulting to QuickLisp over OCICL and API latency clashing with REPL workflows; the author is considering rewriting the project in Go.