As noted by Simon Willison, this small internal observation reveals a major second-order effect. The efficiency gained by automating a request is negated if the recipient perceives it as a dehumanizing delegation. This suggests that for AI agents to succeed in collaborative environments, they may need sophisticated social signaling—perhaps explicitly framing tasks as coming from the human colleague or operating only within pre-negotiated boundaries.
The risk otherwise is that AI, intended to give time back, could instead create new layers of bureaucratic friction and social resentment, undermining the very teamwork it aims to support.