If you’re going to have long-running collaborative AI agent workflows, each instance of the workflow will have to have its own runtime. I just don’t see any way around it. No-one else seems to be hitting this bottleneck, or talking about it.
What am I missing?