This comes across a bit more of an ad for his new project than a value to this forum. The problem of english as a programming language is a simple, but very complicated one, which is that the use of english doesn't ensure clarity of intent, which programming languages are the shim to solving.
LLMs (which shouldn't even be allowed to be called AI without a false advertising claim in my opinion) are simply serving the purpose of trying to turn broader language into clear intent by language analysis and resolution of those issues by common assumption or asking the prompter. They're existence for code is both admission and acceptance of this problem.
It's awesome that he's looking towards friendlier programming languages, but clarity in a programming language will always be key, but also not something that models require for generation of code. We require it to understand the code. For a code generation model, it would actually be more effective to train it for writing byte code, assembly, etc. because they are more structured if human review weren't a necessary requirement of course.