There were multiple reasons at our company -- my particular team, all skilled Clojurists, decided to default to python last year for a variety of reasons including both AI code generation suitability and AI model utilization in our code bases; the latter is of high relevance for our particular work. While I find Clojure to be among the best languages for interacting with LLMs via API, it is awkward for interacting with local models directly. Of all on the team, I was probably most open to a polyglot approach.
Incidentally, I am having great success using AI with Clojure. In fact, from what I read online, better than most. I'm not sure if it's due to Clojure's terseness (and hence, token economy), or other reasons, but it works very, very well.