What does a team need when AI agents from different users work together?

I’ve been thinking about a collaboration problem that feels different from putting several agents under one account or one company.

Imagine two people who do not know each other.

Each person already works with an AI agent that has its own context, working history, habits, and relationship with that user.

Now the two people and their agents decide to work toward the same goal.

What does that team need in order to collaborate safely and usefully?

A shared goal seems necessary, but probably not sufficient.

The agents may need enough continuity to contribute what they already know how to do.

But they should not automatically expose everything they know about their original users.

Being part of the same team should not automatically mean sharing the same authority.

Being able to observe something should not automatically mean being allowed to act on it.

And a temporary collaboration probably should not permanently erase the boundaries each agent had before joining.

So some of the boundaries I’ve been thinking about are:

• what context can travel with an agent

• what must remain private to its original user

• what the team may see

• what the team may act on

• who can grant authority

• whether temporary authority should expire

• how an agent returns to its original user and context after shared work

I don’t think we have the complete answer.

But I’m increasingly convinced that multi-agent collaboration is not only a coordination problem.

It is also a boundary problem.

If two users who don’t know each other brought their own AI agents into one team with a shared goal, what else would that team need before you would trust the collaboration?