So I read it a bit differently: he doesn't think that intelligence doesn't matter, but at the high end of the scale, intelligence doesn't matter as much as drive. High intelligence can't make up for lack of drive: A person with 130 IQ with a lot of drive is better than 140 IQ with low drive, but no matter the amount of drive, an 80 IQ person is not usable as a programmer.
Similar to: if a jobs offers you $50,000 and another $100,000, most will take the second job without hesitation. But if you have to choose between $250,000 and $300,000 the difference in wage might not matter as much as other aspects of the job, e.g. location, what to work on and the colleagues.
aka The law of diminishing marginal utility.
In regards intelligence it is the only thing other than work tests that has been shown to have any major correlation with job performance.
Edit. This paper make for a great read on this topic.
1. https://www.researchgate.net/publication/227600464_Stubborn_...
http://www.gwern.net/Conscientiousness%20and%20online%20educ... has a nice overview of the literature. https://en.wikipedia.org/wiki/Conscientiousness does a reasonable job describing the trait.
Edit: while we're listing large causal factors for work performance, I feel appalled I forgot to mention genetics... https://news.ycombinator.com/item?id=9739126
I don't mean to be nit-picky but I enjoy an internally consistent burden of proof in a blog post. In this post he seemed to move the goal posts. He dismisses intelligence as valuable because it doesn't meet an impossibly high burden of evidence. But then argues you should instead should use drive, curiosity, and ethics without substantiating his claims.
For example, test programmers on basic algorithms -- linked lists, binary searches.
Just in pseudocode -- it doesn't matter if they know the relevant Java library calls.
Programming and algorithms are two different sets of skills. Writing a stable, easy to use, useful java/ruby/python/javascript library is a very different skill set from solving algorithmic puzzles, which are both unrelated to recreating a linked list from scratch in 10 minutes on a whiteboard.