When we make decisions based on AI output volume alone, ignoring everything else that goes on during product development, we do more harm than good, to the point where the organization may be better off abandoning AI development entirely. You cannot assess AI "productivity improvements" without looking at the entire product development flow, taken from when it's a gleam in somebody's eye to delivery to customers and beyond. A simple output-volume metric is worthless on its own. You have to consider the entire system. Goldratt's "The Goal," which I think of as essential reading—knowing that material is a basic software-engineering competence at the level of knowing how to code—addresses exactly this issue. Goldratt talks about robotic automation in a factory damaging overall productivity, but that's exactly the problem we're dealing with when we use robotic (AI) code generators. Automating a sub-process does nothing but jack up operating expenses unless it addresses a bottleneck in the overall development flow. Coding has never been the bottleneck. AI, as often used, does nothing but increase inventory (work done that generates no revenue), and as the inventory piles up, it encourages shortcuts (e.g., inadequate review and testing) that disrupt the flow and damage overall productivity. People do say that "AI creates challenges," but we have not even scratched the surface of how to address those. In the meantime, we make knee-jerk adjustments to our process that damage overall productivity, and given that most organizations are incapable of accurate self-assessment, that goes entirely unnoticed. There are things we can do to get the situation under control. For example: human-created software architecture can be used as a constraint on the AI, working in very small batches makes reviews possible, making the feedback loops faster increases the value of the software and makes necessary changes easier. We can analyze literally everything that happens as a product is developed (the "value stream") and optimize the entire process. We can improve development flow to eliminate dependencies and bottlenecks. We, as an industry, don't seem willing to do any of that, however. We ignore those things at our own peril.