A Workflow Done in Reverse
In the mainstream narrative, AI coding's selling point is speed, but the author deliberately uses it slowly: after writing code, they have the model pick it apart, press on edge cases, and demand it explain every design trade-off—like an indefatigable senior reviewer sitting beside you. Output speed drops, but so do rework rate and production incidents. The author's line is quite apt—the old bottleneck was typing speed, the new bottleneck is thinking quality, and AI can amplify both ends, depending on which you push toward.
This Usage Is Scenario-Dependent
It clearly isn't universal. For prototype validation and one-off scripts where "fast is right," doing it the author's way is pure waste; it suits long-lived core code that's expensive to change. There's an interesting observation in the comments: the most senior people on a team tend to spontaneously arrive at this mode, while newcomers default to treating AI as a ghostwriter—so the tool amplifies the habit gap the two groups already had. How to make the former usage a team default is probably more important than agonizing over which model to pick.
via: Hacker News