Predictable Agents Reorganize the Team Around Them

Most teams about to add an automation will budget for the tool. Few will budget for what the tool quietly reorganizes: who recognizes whom, who routes around whom, and which conflicts spread. A predictable agent does not change a team by doing more work. It changes the team by becoming a stable thing the humans coordinate around.

That is the throughline of the arXiv preprint "When Bots Join the Team," a study of 2,991 GitHub projects watched for two years on either side of each one's first bot. After adoption, the projects showed more repeated collaboration, sharper recognition of who-does-what, fewer conflict cascades, and more distinctive outputs. The changes clustered at the moment of bot adoption, not gradually, and they tracked social organization rather than the bot's own capability.

The preprint's own line, that "the bot is the occasion; social organization is the mechanism," is the part that generalizes. The bot's coding skill is the wrong variable. Recognition, repeated engagement, and role differentiation are the right ones, properties of the team rather than the tool. The "When Bots Join the Team" authors' honest caveat, that the study tracks "precisely timed associations, not causal effects" because there is no untreated comparison group, is the part that keeps the frame from drifting into hype.

A leader rolling out the next predictable agent should optimize how humans and bots learn to route around each other, not what the agent can do. The capability demo is the smaller part of the rollout.