The problem I have with the AI mastering is it seems to be a bit clumsy with making the master fit the song. A lot of these tools just have simple buttons like "low, medium high", and seem to default to really squashing the dynamics to achieve maximum loudness. I have worked with both human mastering engineers and AI mastering, and I would say a decent human mastering engineer is noticeably better at feeling out the songs and finding the right balance.
To your question though, it's a chatbot so tell it you want your track to sound like eg Polo and Pan, it'll look for ways to enhance it in their style rather than a simple small/medium/large knob.
It has powerful MIDI clip editing capabilities via a custom DSL and more recently automation envelope support was added. It also has solid support for the native devices with VST/AU plug-in support coming soon. Basically every aspect of the Live API and Python remote scripts has been exposed via MCP and REST APIs (so you don't even need to use it with AI).
Many people get upset by the idea of MIDI generation, but you can use it however you want. I intended it to be an assistant to help with the tedious and repetitive stuff and to help overcome writer's block.