A lot of small startups are trying to automate timeline construction with VLMs. I've counted about a dozen, some with seed stage funding. If you can crack this, there's absolutely a path to a unicorn. But in the long term, generation will disrupt nonlinear video editors, because whatever high level software we build will be able to dispatch to both generation and editing tasks.

Fwiw, I also work in this space and spend a lot of time thinking about it. Prior to AI, I also spent a significant amount of time making films the old fashioned way.