Lines of code got a better publicist

That framing landed because a lot of people have seen the same pattern inside companies. Leaders reach for code volume, PR counts, and token usage because they are easy numbers to track. They are also exactly the kind of numbers that get gamed once they become targets. Several people tied this directly to Goodhart’s Law and to older management mistakes like measuring hours in seats or raw commit volume. The practical bottleneck, in their view, is not typing code. It is deciding what to build, reviewing it, testing it, and making sure it solves a real user problem without blowing up future maintenance. The strongest thread running through the comments was that AI coding gains are real but narrower than the marketing suggests. Many people said these tools help with prototypes, test generation, code reading, and implementation throughput for bounded tasks. What they do not erase are the serial constraints that matter at team scale: architecture, prioritization, code review, QA, rollout, and business validation. That is why large claims about layoffs driven by AI productivity rang hollow. Most readers saw those announcements as cover for over-hiring corrections, investor theater, or simple cost cutting. If a company truly got a free engineering multiplier, the obvious move would be to ship more customer value and show it in revenue, retention, reliability, or reduced incidents. A second theme was cost. Some people said the tone has already shifted as companies hit token budgets and discover that infinite code generation creates a very non-infinite bill, plus a review burden that humans still have to absorb. Others added that a lot of the new bulk is tests, which can be useful if they are real tests and not generated ceremony. Even there, the useful metric was human understanding. Once the organization no longer knows why code exists or cannot confidently review it, the volume stops being an asset. There was some pushback. A few commenters argued that raw code output is not always nonsense if paired with completion rates, defect data, or constrained use cases like large-scale automated porting. Others said AI is already changing individual workflows enough that refusing to use it is becoming a career risk even if the business case is still fuzzy. But the center of gravity was clear: measuring AI success by how much code appears is managerial regression dressed up as innovation.