I know it's doable because the technologies being used to create Tau for the most part have been done before. The only truly novel part is the TML part which I have never seen before but that is also the part which Ohad is working on right now. It means if he can just build that then the rest is easy in comparison because it is already known how to do the other stuff. Cyc already exists and so people know how to recreate that. RDF? It exists and people like Tim Burners-Lee are working on semantic projects like Solid so again the know-how is there. Logic based AI? It is actually older and more well known than the deep learning stuff. Program synthesis? This isn't new either and neither is formal verification.
So all of it is very doable if TML itself can be made. And the moment we have TML then in my opinion we are safe. Once there is TML and it works then the alpha will be doable and we can collaboratively build out Agoras and everything else.