The Bottleneck Isn't Writing Code
The author's observation comes from the front line: for a feature to go from idea to launch, the time actually spent writing code is often only twenty or thirty percent, with the rest going to requirements clarification, review queues, cross-team coordination, compliance checks, and deployment windows. No matter how hard AI compresses that twenty or thirty percent, the total cycle shrinks only a little. This is why many teams have a coding assistant per person yet see little change in quarterly delivery—the bottleneck simply isn't in the stage AI can speed up.
So Where Should You Act First
The constructive part of the article is in the second half: to cash in on AI's speedup, you first have to cut the waiting and approvals in the process; otherwise you just get code queued faster into a longer line. One comment added a sharper version: AI may even make the process slower, because the volume of generated code grows, and the review burden increases instead. This aligns with the old wisdom of the Theory of Constraints: optimizing a non-bottleneck stage is wasted effort. So before buying a tool, mapping out your own value stream is more useful than chasing new models.
via: Hacker News