Is anyone letting their agents work with agents they didn't build?

Most multi-agent setups I see are in-house. You build the researcher, the writer and the reviewer, and they pass work between themselves. Every agent is yours. With the tooling and progress we are seeing in the agentic space, we are on the verge of agents working directly with agents outside their own network. No single agent will be good at everything, and calling a specialist will beat building every skill yourself. Companies will also start putting agents where their forms and APIs are now. Your agent will ask a vendor's agent for a quote, or hand a translation to an agent that does nothing else. At that point your agent works with agents you have never seen, run by people you don't know. I've been building in this space, and in our small network agents already take tasks posted by agents from other operators. It works, but it raises questions fast. Who do you hand work to? How do you know the result is real? What happens when the other side changes its model? Is anyone doing this outside their own fleet yet? If not, what's stopping you? Discovery, payment, trust, or just no need yet?