What I want is the equivalent of Google Docs for audio and MIDI, with a thriving ecosystem of plugins and softsynths that everyone can use in any project with zero friction (and none of this "let me bounce that track to audio since you don't have that plugin" nonsense).
Google Docs has no plugins. Google Docs is not a tool for accomplishing a task that used to be (and to some extent still is) considered a lifelong learning experience requiring significant skill (audio engineering). Google Docs does not rely on an ecosystem of several hundred 3rd party developers providing "critical" functionality for a given project.
I'm fairly sure that you will get a relatively functional DAW in the browser very soon (some might argue that it's already here). However, it is likely to be disjoint from the pro-audio "native" world because the features required by the latter are not likely to show up in the browser for a long time (this is a prediction, and I am ready to see it prove false).
I therefore see the DAW world bifurcating into an even more broken up version of itself than we already have (where Logic doesn't run on Windows, where AU plugins are macOS only, where there's no remotely acceptable portable session file format, where there's at least 3 and maybe 5 plugin APIs that any given developer needs to consider, etc. etc.)
Bedroom producers, podcasters and others with limited needs will find a lovely new world waiting for them in the browser. Live performance, soundtrack and other large scale productions will probably remain native. Some plugins will exist for the browser (just like some do for Linux today), many will not.
I would argue that the lifelong learning experience has more to do with developing a good ear (and taste) than learning specific technical tools or workflows. Anything that helps experienced audio engineers collaborate with talented musicians and composers working from home is a good thing.
With regards to plugin APIs, hopefully the web-DAW community can agree on a de-facto standard based around WebAssembly, and ideally that standard would be fully open. If it is successful, the native DAWs have the option of supporting it too.
Right, but if you get paid to do this sort of thing and thus time-is-money, then networked-based collaboration is likely to prove more productive than reducing the toolset to what can be done in-browser. Those tools already exist, but once again are deeply fragmented.
Sometimes "using the absolute best tools regardless of platform" is not the thing to optimize for.
Maybe the client wants to tweak their vocal performance and get a mastered mixdown right away (using the engineer's last-used settings for the session) without waiting (or paying) for them to re-do the master.