In my opinion what makes REST REST, is its set of verbs and error codes and their attributed semantic meaning and the documented client and server behaviour for each of these . What the author is complaining about, which seems valid, is the over fitting of use cases to fit this ‘golden set of verbs and http codes’. The self-describability looks like a useful feature well implemented on REST but can as well be implemented on any protocol as you have agreed. But your statement that building a self-describable system will make it much similar to REST is not accurate.
What i will grant is that REST implementations come with a lot of good auxiliary features like headers and cookies and self-describability and standard referenceable schemas, which no doubt should be appreciated and adopted, but the basic essence of the author’s argument that it is time to move on from REST is still worth appreciating and thinking about.