I think they're referring to the problem of measuring programmer output (and thus, measuring programmer efficiency). And how the output of that work (lines of code) can actually be reduced, but still considered productive.

I'm not putting this very well.

In most areas of business, there are important signals which can be observed that are often based on counting values over time. There are people for whom it is their primary job function to identify what should be counted, and how.

When a codebase shrinks AND increases in quality, you've now demonstrated that the most obvious way to quantify the utility of a resource - units of productive output - is not useful. So now people who are used to tallying widgets and graphs that are "good" if they move up and to the right, now have to understand things like refactoring, technical debt and design patterns.