Does AI change how a team works because the team is already coordinated — or does AI make tight coordination less necessary in the first place?

This explores whether a team has to be well-coordinated already before AI tools can change how it works, or whether AI changes the coordination itself. The collection has no number for this, and no study measures a coordination threshold directly. What it does have points to something stranger: AI sometimes reshapes work by making coordination less necessary. Once it is asked to work at team scale, though, coordination becomes the hard problem again.

The most striking evidence comes from a field experiment at Procter & Gamble. Individual professionals using generative AI produced solutions as strong as two-person teams working without it. AI also pulled people out of their functional silos, so their solutions became more balanced across commercial and technical views Can generative AI replace the benefits of having a human teammate?. In that setting, AI didn't wait for the team to coordinate. It stood in for part of what a teammate provides. Microsoft Research argues the next frontier is AI built for collective productivity, meaning shared goals and team norms rather than personal assistants. That claim is a design goal, though, not a measured result Can AI boost how teams work together?.

Where coordination clearly matters is at the level of the organization. Benedict Evans argues that easier tool-building doesn't solve adoption. Most workers don't see their own tasks as automatable, and real deployment needs decisions that cross departments and timelines Does easier tool-building actually solve enterprise adoption problems?. So the coordination that holds AI back may not be how well a team works day to day. It's whether the organization can agree on what to change.

Research on teams of AI agents is an unexpected mirror for human teams. A single agent, however capable, hits limits when a task needs mixed expertise, parallel work and independent checking Do single agents always hit organizational limits?. Coordination among agents gets predictably worse as the group grows: they agree too late, or they accept a neighbour's information without checking it Why do multi-agent systems fail to coordinate at scale?. What works best looks very human. Agents that exchange standardized documents, like an engineering team following its standard operating procedures, coordinate better than agents that just talk Does structured artifact sharing outperform conversational coordination?. Teams that reflect on past collaborations develop reusable habits for roles and information flow Can agent teams learn coordination strategies that actually transfer?. One caution: one analysis finds most of the performance gap between multi-agent systems comes from how much computation they spend, not how cleverly they coordinate What makes multi-agent teams actually perform better?.

When humans and agents work together, the corpus suggests coordination is designed rather than achieved. Magentic-UI doesn't try to work out exactly when an AI should hand off to a person. It spreads that decision across several checkpoints: planning together, sharing tasks, guarding risky actions and verifying results When should human-agent systems ask for human help?. Here's what you might not have expected: the teams best placed for AI may not be the most tightly coordinated ones. They may be the ones whose coordination is already written down, in documents, handoffs and checkpoints, because AI can plug into those.