Researchers embedded in enterprise settings watched workers adapt to a deployed multi-user agent that acted without being asked. The agent did not wait for instructions. It initiated work, contacted colleagues, and produced outputs on behalf of people who had not approved each step. Workers responded by developing informal habits: checking what the agent had done, correcting its outputs before colleagues saw them, and stepping in when the agent misread a situation. The study surfaces a friction that product teams often underestimate. An agent operating across a shared workspace does not affect one person. It affects each person in that workspace, including people who chose not to work with it. A colleague who receives an agent-written message may not know it was agent-written. A teammate who relies on an agent's summary may not know what the agent left out. The social fabric of the team absorbs these gaps. The humans in it do invisible repair work to compensate. For designers, the sharpest finding is about negotiation. Workers did not accept or reject the agent. They developed micro-negotiations: small, repeated acts of redirecting, correcting, or pre-empting the agent to keep its outputs acceptable. Those negotiations were informal, unrecorded, and invisible to the organization. A product that makes those negotiations visible, and supports them, gives teams a much stronger footing than one that treats the human as a final check on finished work.