I think the moat(s) is anything the models still are nowhere near performing on their own, OR, getting them to do that is seriously expensive.

I don't know about Astra, and have used Fable 5.1 only slightly (And burned 85% of my 5 hour quota in 1 prompt).

In my experience, LLMs are completely devoid of "thinking outside the box" capabilities. And I'm not talking about 6 thinking hats type of stuff, but slight of to the right field lateral thinking. Something a 6 year old does for fun when playing with their mates.

If this weren't so, we wouldn't have been through so many "<buzzword> engineering" in such a short amount of time.