Other things that make it difficult, some mentioned in the article: - It's a strange activity. The output is non-linear, non-repeatable, and basically chaotic

Other things that make it difficult, some mentioned in the article: - It's a strange activity. The output is non-linear, non-repeatable, and basically chaotic in some ways. - You almost never build the same thing twice. Even if you do, productivity is affected (positively or negatively) just by the fact that you already did it before. - The value of what you produce is unknown or fluctuating wildly. You might be a unicorn one day and bankrupt (even personally liable!) the next. Less extreme examples are interesting too, maybe your software is a plugin for a product that stops supporting third party plugins. I think, like the article says, that pretending to be able to measure output just because we have something that might look superficially like actual output in other activities (e.g. number of units produced in a factory) is fundamentally misguided. It's a cop-out since what's really missing a lot of the time is (good) leadership.