June 24, 2026
The bot wrote a novel. The bill screamed
What I'm Finding About LLM Code Style and Token Costs
People are roasting AI for writing bloated code that runs up the bill
TLDR: The article argues that AI coding tools waste money by producing overly long answers instead of using simpler built-in web features. In the comments, users agreed hard, sharing cost-cutting tricks, roast-level horror stories, and a growing belief that AI only works well if humans tightly control it.
A developer went looking for a simple answer to a painful question: why is AI-written code so long, messy, and expensive? Their big claim is that the real money drain isn’t what you type into the chatbot, but the huge walls of code it spits back out. In plain English: the robot keeps solving old problems with extra fluff, even when today’s web tools already have shorter, built-in ways to do the job. And yes, the author even drags the ancient tabs vs. spaces argument into the chaos, basically suggesting that every extra character is now a tiny tax on your wallet.
But the real fireworks are in the comments, where readers turned the post into a group therapy session for people exhausted by overexplaining AI. One fan called the piece “impeccable analysis” and said it would change how they work. Another backed it up with hard numbers, claiming that forcing replies into strict JSON cut output by around 40% because the model stopped rambling like a nervous intern. Others were less polite: one user said Claude nearly turned an RSS feed into “ZALGΌ IS TOƝȳ THË PO NY” before being reminded a simpler option already existed. Ouch.
The hottest consensus? AI is useful, but only if you already know what the right shape of the answer should be. Otherwise, commenters say, it happily burns your money generating verbose comments, fragile code, and pure confusion. The mood is part practical advice, part roast battle, and part warning: if you let the machine freestyle, don’t be shocked when your invoice starts looking scarier than the code.
Key Points
- •The article says LLM-generated code can drive high API costs because output tokens are several times more expensive than input tokens.
- •It argues that generated code often uses longer, more fragile, and less secure patterns instead of capabilities already built into the web platform.
- •The author says code style and formatting choices can have practical economic effects because they change token consumption.
- •The article highlights Deno and Cloudflare Workers as runtimes that natively expose browser-style Web APIs on the server.
- •It links WinterCG’s work on a common runtime API surface to reduced need for shims, translation layers, and extra generated code.