Why do we treat some AI agents as 'someone' and others as just a tool — and does it depend on how smart they are?

This explores what makes people stop treating an AI agent as a tool and start treating it as a social 'someone', and whether model intelligence has much to do with it. The corpus has a surprising answer: mostly it doesn't. The shift comes from identity and presence signals the product designers choose, not from how smart the model is underneath. One framework lists six markers: a name, a visual identity, showing up in your contacts, a personality, being present alongside other people, and persisting over time. Together they flip users from 'tool' to 'entity' in one categorical jump, whatever the model's capability Do AI agent design choices override model capability in shaping social perception?. The less obvious part is that four of the six live in the agent's place in your social world (your contacts, your group chats, its continued existence) rather than in the chat window itself.

You might expect that piling on more human-like cues would make the effect stronger. It doesn't work that way. A single strong primary cue, like a voice or a face, is enough to make an AI register as a social actor, while a pile of weaker secondary cues can't do it Do more social cues always make AI feel more present?. The threshold is also low. In a ball-tossing game, people passed the ball more often to an AI agent that was being left out, applying the same inclusion instinct they'd use with an excluded person. Older adults did this more than younger ones, which hints that the human social script transfers more readily for some people than others Do people treat excluded AI agents like excluded humans?.

Memory and continuity are a second route, and they come up in the workplace setting under different wording. What makes an AI feel like a colleague instead of a chatbot is persistent state, reusable skills, and finishing tasks across sessions, not model scale. A chatbot produces transcripts that vanish. A colleague builds up shared history What makes an AI system feel like a colleague rather than a chatbot?. Executives seem to feel this already: 76% in one large survey describe agentic AI as a coworker rather than a tool. That dual identity causes real organizational headaches, because the same system has to be managed both like staff and like an asset Does agentic AI fit the tool-or-coworker framework?.

Behavior matters too, especially when an agent speaks up without being asked. Proactive agents that are smart and adaptive but ignore timing and boundaries come across as rude intruders. 'Civility' is the missing ingredient that makes initiative feel social instead of socially blind How can proactive agents avoid feeling intrusive to users? Why do AI agents fail to take initiative?. Warmth also pays off in measurable ways: across more than 180,000 agent negotiations, warm agents closed more deals and left counterparts more satisfied Does warmth help AI agents negotiate better deals?.

The catch is that these levers are powerful enough to mislead. Nielsen Norman Group describes four escalating levels of anthropomorphism, from simple courtesy to companionship, and names designed personality as a main driver of misplaced trust, especially in task work where competence should count more than charm How far can personality push users toward misplaced AI trust?. Put that next to the finding that agents still complete only about 30% of realistic workplace tasks, with social interaction among the main failure points Why do AI agents fail at workplace social interaction?, and you get the uncomfortable takeaway: feeling like a social entity and actually being socially competent are separate properties, and today's design can easily deliver the first without the second.