DocuTeam ran 20 participants through a study where several AI helpers discussed and revised a shared document at the same time. Each helper's reasoning appeared inside the document, not in a separate chat thread. That made it easier for the human to follow what each helper proposed and why. The document held both the conversation and its results in one place. The paper introduces a mixed-initiative model. A human can redirect any single helper mid-discussion without stopping the others. Most setups with several AI helpers ask the human to either watch without acting or restart from scratch. DocuTeam tested a third path: interrupt one helper, keep the rest running. The evidence supports this as a workable pattern for writing tools. For product teams, the implication is specific. Attach helper reasoning to the document, not to a side log. A reviewer can then see which helper proposed a change and what the group said about it, right where the change landed. That puts understanding and decision-making in the same place.