I'm willing to bet it's because model providers have gone all-in on programming and reasoning ability when training new models. It's where the money is, if nothing else, and the writing quality will have suffered as a knock on effect.