I don’t disagree, but I think you underestimate the importance of the medium for collaborating with others. If you design a system with a high barrier to adoption (e.g. learning curve), you will remain its only user.
You should, however, think about the foundations before choosing the medium. Otherwise you will make something popular but long-term toxic. We already have too much of that.
What I’m saying is that the medium shouldn’t tightly coupled with the foundation. The foundation should be able to reason about medium it speaks in, while the mediums should be an extensions of the foundation. We have that with natural languages and thoughts, where our brain as foundation can manifest thought into any medium, while beimg able to adapt under new mediums.
So ideally foundation should do the same: be open to new mediums and convert “thoughts” between them. I don’t see why this model is toxic in any way.
EDIT: There is a rabbit hole of a problem, that won’t fit nicely within this post. We have Racket that can cast representation between medium, but ultimately it’s ecosystem is shattered, because it doesn’t know how these mediums interact. It literally about building thousands of beautiful towers that can’t brisge between eachother.
If everyone sets up their own foundation and builds media on top of it, we can’t communicate any more!
In practice we need to make compromises. Building various systems and bridges between them. Some of these systems are messy legacy systems (like the Web standards). That’s life.
Done. I saved the notebook again with a different title, then went to the ledger to delete the old one. All that works just fine.
It looks perfectly possible to work with nothing but PDS storage, which could be useful e.g. on a mobile device.
Now I can start to think about a bridge between Lopecode and HyperDoc 
I wasn’t saying about building more foundations either, I was talking about building 1 ultimate foundation. This one is hard to do because you’ll have to pick correct constraints and try to make the logic work around those constraints.
For example I spent like 3 years just designing NegI, without even trying to make an implementation, because it wasn’t about simplicity of use, but rather simplicity of rules and guarantees that create emergent and complex interactions. While I don’t know the Rust and Erlang, I did realise that their popularity comes because of guarantees, rather than simplicity of use.
Okay, all the links broke for something to do with PDS garbage collection that I do not fully understand at this point… will try to fix!
edit: FIXED, was misconfigured cloudflare caching, not ATProto, I am trying to avoid hammering PDS with intermediate caching but I was caching transient 4xx requests.