Is anyone else terrified of their codebase right now? My team’s “velocity” is up 40% thanks to LLM copilots, but half the new code feels like highly optimized technical debt. We’re shipping faster, but I spend more time debating if the AI’s solution is correct or just plausible. What metrics do you trust besides commit counts?
Velocity is a vanity metric, always was. Leadership loves it because it moves up and to the right. The true cost shows up 6 months later in maintenance P0s. The LLM just accelerated the initial illusion.
We flipped our metric. It’s not code lines, it’s Cycle Time to Value. If the AI saves 3 hours on a feature, that’s 3 hours the engineer can spend on architecture review or user research. The PMM view is all that matters, is the feature driving activation?
Welcome to the new era. Your job isn’t writing code; it’s being a human input validation layer for the $100M-funded hallucination machine. Your company will measure lines of code until the inevitable re-write, then blame the ‘underperforming’ developers