While I see the merit in his arguments, I believe his approach to data is "Clojure-like" or Lisp-like, in that it discourages explicit enumeration of all states and their possible configurations for the tradeoff of being able to sculpt these records. But, as a Haskell dev, I do not want to sculpt the records in the way he describes. I want to explicitly enumerate all the possibilities. This is, in my mind, a difference in cognitive preference. There is no right or wrong way and I think what he proposes is elegant. It is ultimately a matter of up front reasoning about the space compared with reaching a synthesis of the problem domain over time through wrestling with the problem itself. It is a statement about the efficacy of saying what can be known up front and to what extent, how much accuracy, and what utility.
I am about to launch a SaaS that pairs languages to these cognitive patterns as explicit understanding of such information can help bring together people who think similarly about information (team-building) while also opening up the possibility of expanding ones insights with alternative ideas and approaches (as his talk has done for me). The intent is to help hiring teams find devs that match their culture's programming and problem solving styles (whether that be reinforcing or doubling down on ways of thinking and acting).