The Metric Trap: LOC Is Back, Now With AI Spin

A post arguing that 'lines of code got a better publicist' landed on HN's front page and sparked a thread about how AI is resurrecting the oldest bad metric in software. The core argument: before AI, the goal was to make money, with AI the goal became to ship more code, and the right post-AI goal should still be to make money. One commenter tied this directly to a survey where Augment asked 219 engineering leaders to define 'AI-native engineering' and got 219 different answers, which landed as a punchline but also as a diagnosis.

The pattern: as AI coding tools make it cheap to generate code, the gravitational pull toward measuring output by volume increases. This is exactly backwards. The value of AI coding assistance is not more code, it is faster delivery of working software that solves real problems. When the tool makes code cheap to produce, the temptation is to treat code production as the output metric, which is how you end up with a Claude agent spending thousands of tokens to fix two lines of CSS and calling it productivity.

This connects to the broader frustration in the 'software is made between commits' thread, where one commenter noted that good software tools should be invisible, like a hammer, and that constantly thinking about your tools means something is wrong with them.