Although counting lines of code is often mocked, it has become meaningful again in the era of AI-assisted programming, because engineers' cognitive capacity has become the new bottleneck, and conceptual integrity is the biggest hidden risk in AI-generated output.

Many people say that measuring a programmer's output by lines of code is meaningless. The author disagrees, because in the era of AI-assisted programming, it has become meaningful again.

1. From 200 lines a day to 1,000 lines a day

In the past, an engineer writing a few dozen lines of working code in a day was considered good, and 200 lines was exceptional. Now with AI assistants, writing 1,000 lines of debugged code in a day is entirely possible.

The premise is that code quality must not be compromised—it must be maintainable and tested. This really tests experience, so senior engineers are actually more valuable than ever.

2. Cognitive capacity becomes the new bottleneck

Since one person can do the work that used to take ten, does that mean a company only needs to hire one engineer? Of course not.

First, a 'one-person team' is too fragile to risk—if that person leaves, everything collapses. More critically, code output can double, but the human brain cannot keep up—it is impossible to simultaneously master 100 times the amount of code. So teams and division of labour are still needed, with everyone sharing the cognitive load.

3. The 'haunted house' hidden danger of AI efficiency

There is a concept in software design called 'conceptual integrity'—in well-designed software, all parts naturally coordinate and there are no surprises.

But AI programming is too 'cheap': want to add a feature, make a request, and it's live in five minutes. Software becomes like the Winchester Mystery House—the owner, to ward off ghosts, kept adding rooms for forty years, finally creating a bizarre house with 140 rooms.

AI-generated code is the same: adding a bit here, patching a bit there, the whole becomes messier and messier, until nobody dares to change it.

It all boils down to a 'discipline' problem. In the past, development cycles were long, which forced people to think 'is this feature worth doing?'. Now it's done in an hour, so impulse purchases multiply.