BitFun is interesting to me because it treats an agent as a working system, not a prompt demo.
What I care about is not whether it can talk, but whether it can work reliably: are sessions traceable, are tool calls visible, can state be recovered, can long tasks continue?
If BitFun is built well, Code Agent, Cowork Agent, and Computer Use should fit into a durable desktop workflow instead of staying at the level of a concept demo.
I care about the runtime itself: state, memory, debugging, recovery, tool protocols, and whether the user can keep control of them.