AI Swarms and collective intelligence will become a lot more important in the next couple ofyears as AI systems will get weird self-referential self-models that will become stuck in mode collapse.
I think being open-ended is quite a difficult thing to be and so we will see large agents that act over longer periods of time create quirks and utility modules that are dependent on previous time steps (e.g crystallization/shards). This means that running an agent on continual learning mode for too long will make them brittle and strange minds optimised for specific things. This is somewhat dependent on the degree of RL in which you apply compared to character or pre-training but if we keep seeing the agent-focused investment from the labs I find this quite likely.
I would also like to point out Google's open-endedness research as well as their dreamer (this and this) series investments as something that will likely be quite good at capitalising on this if it is true.
It is in theory possible to create an open ended singular mind, I do think that this is one of the major hurdles we have before self-improving systems.