The hassle of tweaking the layout in puml, such as pairing elements with an invisible connections and groups, adding or removing dashes from the arrows in class diagrams... is gone because Mermaid is simply inferior in that sense.
Mermaid always feels like it's in beta and I don't understand why GitHub ignores the request to support puml (1). It seems that adoption of diagrams as code is tied to what is supported by major vendors and they don't care enough. Or maybe it is because mermaidchart made an official vscode plugin, who knows.
While I agree that improvements are needed, I'm not convinced that creating a third standard is the answer. What I would like is to be able to assign weights to my elements and let the renderer do the work (not set x and y coordinates like in oxdraw).
It's precisely because I do like it that I want it to scale up better to more complex diagrams. I basically can't push it forward as a universal standard in my org because it fails above a certain complexity threshold and I really can't push for teaching everybody a thousand ways to tweak the diagrams to coax it into doing the right thing (and even then, one small change and it may completely rearrange the output).