Being good with AI may be its own skill, separate from being good at the job, and tied to reading what the AI knows.

This explores whether perspective-taking, the everyday skill of modeling what someone else knows, wants, and is likely to misunderstand, explains who gets the most out of working with AI. The corpus's most direct answer is surprising. The people who collaborate best with AI are not simply the people who are best at the task. In one study, users with stronger theory of mind got better results when working with an AI, but they had no advantage when working alone Does theory of mind predict who thrives in AI collaboration?. Being good with an AI looks like its own skill, separate from being good at the job, and it is closer to being good with people than to being an expert.

The more useful detail is that this is not only a fixed trait. The same research tracks perspective-taking from moment to moment within a single conversation. When a user is actively thinking about what the model knows and needs at a given turn, the AI's next response measurably improves What breaks when humans and AI models misunderstand each other?. That note also raises the stakes: the human and the AI each hold a model of the other, and when those models drift apart, the result is not just awkward conversation. It can lead to the AI taking the wrong action on its own. Perspective-taking works less like a talent you either have or lack and more like ongoing maintenance of a shared understanding.

This changes how to read nearby findings. People's mental models of AI partners are mostly built on perceived competence, with human-likeness and conversational flexibility coming after How do users mentally model dialogue agent partners?. A good perspective-taker presumably keeps those impressions tuned to what the system can actually do. Over repeated interaction, people do update: in partner-selection games, people overcame an initial bias against AI and learned to prefer AI partners because the bots behaved more reliably Do humans learn to prefer AI partners over time?. The other side is the 'LLM Fallacy,' where people misjudge their own contribution and credit AI output to their own ability How does AI-assisted work reshape how people see their own abilities?. That is a failure to keep track of where your mind ends and the machine's begins, which is perspective-taking pointed at yourself.

If perspective-taking can change within a conversation, design can support it instead of just selecting for people who already have it. Several notes point this way without using the term. In one study, assistants that asked reflection questions alongside their advice beat assistants that only gave answers Do reflection questions help people make better decisions with AI?. Systems that highlight what to look at, rather than handing over a verdict, reduced anchoring on the AI's answer Can AI guidance reduce anchoring bias better than AI decisions?. Agent interfaces that build in co-planning, check-ins, and verification steps give users regular chances to re-sync their picture of what the agent is doing When should human-agent systems ask for human help?. The reverse direction also matters: proactive AI that anticipates what the user needs is the machine doing its share of perspective-taking, and it is still rare in research datasets Could proactive dialogue make conversations dramatically more efficient?.

One limit is worth knowing. A parallel finding from multi-agent AI teams shows that diversity of perspectives helps only when it rests on real expertise. Without that, it makes the team worse Does cognitive diversity alone improve multi-agent ideation quality?. That is about AI teams, not people, so it is an analogy rather than evidence. But it suggests that perspective-taking multiplies what you bring rather than replacing it. The corpus has strong evidence that theory of mind predicts collaboration gains, and much less on how it interacts with domain expertise or whether training people in it improves results. That gap is an open question.