AI will be both a floor and a ceiling raiser, since there is a practical limit to how many domains one person or team can be expert in, and AI does/will have very strong levels of expertise/competency across a large number of domains and will thus offer significant level-ups in areas where cross-domain synthesis is crucial or where the limits of human working memory and pattern recognition make cross-domain synthesis unlikely to occur.

AI also enables much more efficient early stage idea validation, the point at which ideas/projects are the least anchored in established theory/technique. Thus AI will be a great aid in idea generation and early stage refinement, which is where most novel approaches stall or sit on a shelf as a hobby project because the progenitor doesn't have enough spare time to work through it.