Nathaniel Whittemore argues that AI agents should support the work people do together as well as individual productivity. A shared agent gives teammates access to the same live session, durable project context and work in progress, allowing another person to inspect, redirect or continue the task without reconstructing a private conversation.
Nathaniel Whittemore points to shared agents in Slack and collaborative agent interfaces as examples of this direction. The useful distinction is between sharing a finished answer and participating while an agent works. These product and team examples support his argument for experimentation, but the episode does not present measured results from a completed team pilot.
Nathaniel Whittemore outlines a four-stage approach. First, inventory how teammates currently use AI and identify barriers or internal practitioners who can help. Next, assemble the context the team needs to share. Then map recurring work to see who touches each workflow, how often it runs and which information it depends on.
Nathaniel Whittemore suggests assessing candidate workflows by shared need, the cost of outdated information, permission sensitivity and whether results are easy to check. A first pilot should put one shared agent into real work with at least two teammates using tools they already have. The team then evaluates whether it improved the work and tries another use case if the benefits do not materialize.
Watch on YouTube
