More code, same delivery date.
A development team started working with AI and their output went up fast. Within weeks they were producing far more code than before. On paper, a win. In practice, nothing reached users any sooner.
The testing environment had become the new bottleneck. It was built for the old pace, and now everything queued up in front of it. The team had not gained speed. They had moved the queue. That is what tends to happen when AI lands on work that was divided up years ago: the constraint shifts one step down the line, and the gain quietly disappears into it.
This is where most of the work stops, because the brief said development. We went through the whole cycle instead: what runs where, what waits on what, and which steps only exist because there used to be time to spare. Then we divided the work differently, AI included. That is when the extra output turned into software people could actually use.











