Ask HN on Shipping AI Code You Don't Understand Hits 391 Points

An Ask HN thread opened by user 'rzk' asking how to handle shipping AI-written code engineers don't fully understand drew 391 points and 538 comments in about 15 hours, pointing to a LinkedIn post from a Meta engineer who wrote that "I'm writing code these days and it all feels like a black box...You understand the high level, but not every detail."

The top-voted response, from commenter bitwizeshift, pushes a design-first answer. "I take the time to create the full hierarchy of abstractions and actors that I want my code to do. I decide, at the UML level, how abstractions will sit and interact — and then I make the LLM implement those gaps," they wrote. The claim is that comprehension survives even when individual implementations stay opaque: "I DO know how the code works. I may not know how each unit is implemented, but I know each unit's responsibilities and how they interact." The critique of unguided generation is blunt. "I've found that LLMs are HORRIBLE at writing decent abstractions...It will always get the job done, but it will often be unmaintainable." Upfront design is framed as a trade: more planning hours, fewer tokens burned, better code later.

Commenter jedberg offers a different frame. "With AI, everyone now gets to act like a PE. Your job is to design, the AI builds like the junior engineers," they wrote. "You may never understand the details, and that's ok, as long as you know the main ideas."

Other high-engagement replies, from users including davepeck, Silhouette, swatcoder and birdsongs, run from workflows that sketch a design in Python before a Rust implementation to louder worries about whether this way of working produces maintainable systems and about the professional culture allowing it.

It's one thread on one forum, part of a run that includes our 123 Coding Tools stories of the last 90 days — among them Replit's disputed claim that it had moved off Anthropic for OpenAI.