HCI researchers map how shared workspaces change agent work

A longitudinal study by HCI researchers produced a conceptual model for designing workspaces where humans and agents work side by side. The core finding is that the workspace itself, not the agent alone, carries the knowledge needed to get work done. A malleable workspace, one that both human and agent can reshape, lets the division of labor shift in real time. The research identifies a recurring problem: agents lose the reasoning behind a decision when they only see the output. Humans lose track of what the agent has changed and why. A malleable workspace addresses this by making intermediate states visible. The human can see what the agent is working on, add background before the agent proceeds, and redirect without starting over. For product teams, the practical lesson is that the workspace design determines how much trust a human can extend. A workspace that shows provenance, meaning where each piece of knowledge came from, lets the human calibrate before approving the next step. A workspace that hides intermediate work forces the human to approve without that picture, or check everything from scratch.