THE SHORT ANSWER
Because the company made generation free and never funded the reader. When Claude Code or any generator writes the specs, code, tests, and tickets, producing artifacts stops being the constraint within a week. The constraint becomes reading, reviewing, judging, and knowing whether the thing worked, and that step got no new capacity. Goldratt's rule applies: speed up a step that is not the constraint and you do not go faster, you pile inventory in front of the step that is. In an AI-first org the inventory is unread PRDs, unreviewed diffs, and tickets closed by the same model that opened them. Output goes vertical, outcomes stay flat, and the people pressing enter for twelve hours feel slow because nothing they touch ever lands. The viral post from voxium, reshared by Aakash Gupta, is that pattern described from the inside, including management's own diagnosis: pushing code is not a bottleneck, so why are we slow. The fix is an operating model, not a tooling change: fund review and eval capacity like headcount, name a reader of record for every artifact, put evals before volume, give one Product Builder ownership from prompt to landing, and measure direction weekly instead of tickets closed.
The viral post is a story about Claude Code the way a traffic jam is a story about cars. The cars are fine. The road was designed for a different flow.
Read it as a systems problem and it stops being depressing. Management said the true thing out loud: pushing code is not the bottleneck. They just never asked the follow-up. If it is not code, what is it, and did we give that step any capacity when we made code free?
The answer is in the post itself, in the part where nobody has time to check the code and see what is going where. That is the constraint. It has a name, it can be staffed, and it can be measured. The one rule I would install first is the reader of record: no artifact gets generated until a named person agrees to read it end to end. It kills most of the pile in a week and makes the missing reading capacity visible enough to fund. Orgs that do that keep the tools and lose the twelve-hour days. Orgs that do not will keep asking why they are slow while the output charts climb.