Write code like a human will maintain it

Category: Engineering post
Status: discussion
Sources: unstack.io, HN discussion
Summary: The post argues that code-quality discipline matters more, not less, when a large language model writes the code. Its claim is that LLMs reproduce the patterns already present in a codebase, so merging duplicated or poorly structured AI-generated code trains the assistant to repeat those shortcuts on later requests, compounding the problem. The author recommends holding AI-written code to the same review standard as human-written code rather than assuming a future tool will clean it up.
Comments: HN commenters note the author could extend an agentic code-review prompt to catch the duplication described. Others argue that thorough review of AI output erodes the speed advantage, while some report shipping more features by accepting that tradeoff.
Why it matters: It connects the AI-assisted-rewrite theme running through the week (Bun and pgrust) to day-to-day review discipline, arguing code quality feeds back into future agent output.