I think that there will be ways to achieve this. If you look at many of the current innovations around working with llms and agents, they are largely around co

I think that there will be ways to achieve this. If you look at many of the current innovations around working with llms and agents, they are largely around constraining and tracking context in a structured way. There will likely be emergent patterns for these sorts of things over time, I am implementing my own approach for now with hopefully good abstractions to allow future portability.