AI-generated code makes reviews expensive and rewrites cheap

Reviews have become expensive, rewrites have become cheap

LLMs don't have an instinct to reach for the shortest path—writing 200 lines of implementation costs them the same as two lines of import. The result is technically correct but over-engineered code that makes reviewing expensive: you keep deciding whether to accept complexity or push back. Rewriting, on the other hand, is now cheap—ask the same model to simplify, use a library, or cut unneeded features. The author now spends more time upfront on scope and library choices, deploys to a test environment, spots what can shrink from 100 lines to 10, and rewrites it. Letting complexity through is no longer a sunk cost you have to live with.

Why it matters: A sharp frontline observation from an engineer that captures a specific LLM coding behavior — defaulting to build over import — with direct relevance for AI-assisted coding practitioners. Points off because it's a personal blog post with no data or controlled experiment, and t...