AI Is Forcing Us to Write Good Code

AI agents don’t clean up messes—they smear it. That’s the team’s pitch: if you want bots to write code, you need guardrails like tests for every single line, tiny well‑named files, and fast throwaway dev sandboxes. Cue the boldest claim: 100% code coverage (meaning every line has a test), or chaos.

Then the comments erupted. Supporters like mkozlows cheer that this flips “best practices” for humans into bot fuel—LLMs (large language models) crank code and need clear targets, so full coverage becomes a map. Skeptics swung hard: badgersnake dunked on “blogging engineers,” while mrits slammed the metric as gameable (“write tests, still break everything”). The meme of the day? The author’s own Roomba‑dragging‑dog‑poop analogy—commenters joked the bot needs child locks and paper towels. Meanwhile, jennyholzer3 wonders if LLMs can ever keep up with genuinely new ideas, not just remix old ones. And bgwalter casually drops a Logic link promising “typed, tested, versioned” APIs in minutes—half helpful, half infomercial, all fuel for side‑eye.

It’s a vibe: engineer beekeepers trying to keep their AI hive busy and clean, while the peanut gallery argues whether 100% tests are a safety net or a vanity metric. Drama level: high, coverage level: higher.