I haven't tried it yet but does running SOTA open source models on the cloud providers or some of the newer AI inference providers like together, etc not work well?

I can’t wait for proper support in the protocol to enable hosted MCP servers. Most enterprises wouldn’t want folks to download and run MCP servers locally. The support today feels like an afterthought.

I was excited to see this!

Love paneer!

I think it’s both? Traditionally, an “agent” is viewed as something that can perceive its environment and act on it (thus interact). It would be important for an “AI agent” to display some level of autonomy. Not sure if I would classify one as a more distinguishing feature over the other.