Every big tech company wants you to feel warm toward its AI. Microsoft's new agent Scout drafts your emails and reshuffles your calendar. Meta's Muse, launched in September, comes with a cartoon avatar you can name yourself and has already passed 2.5 million downloads.
A startup called Skydive goes further, giving its agents Muppet-style faces and job-specific names, like Canary, an agent that watches over production systems around the clock. This is a deliberate design choice, not a happy accident. The people building these products have found that a face and a name make an AI easier to trust and easier to keep using.
One startup founder described watching employees start referring to an agent as someone they "worked with" on a project, the same way they would talk about a human colleague. The problem is that this same trust is measurably bad for oversight.
Boston Consulting Group ran an experiment where managers reviewed identical work, told in one case it came from an "AI tool" and in another that it came from an "AI employee." When the work was framed as coming from an employee, managers caught 18 percent fewer errors in it. The label alone changed how carefully people checked the output.
This is not a fringe experiment. BCG's broader survey work found that nearly one in three managers already talk about AI as a teammate or colleague, and more than one in five say their organization has put AI agents on the company org chart.
That number is set to climb as more vendors sell agents by the seat, not the license. There is already a cautionary tale on record: HR software company Lattice announced in 2024 that it would give AI bots official employee records, onboarding, and a spot on the org chart. Within three days, industry backlash forced the company to scrap the plan entirely.
The idea did not disappear, it just moved underground into products that personify agents without calling them employees outright. For a business owner, the practical risk is not that an AI agent makes mistakes; every piece of software does. The risk is that giving it a friendly name and a face makes your own staff less likely to notice when it gets something wrong, precisely when the agent is handling contracts, customer emails, or financial reconciliations without a human reading every line.
The fix is not complicated. Keep calling it a tool in official communication, even if staff privately name it something friendlier. Set explicit review checkpoints for anything an agent produces that touches money, contracts, or customer commitments, and do not let those checkpoints soften just because the output looks polished and arrives with a cheerful thumbs-up emoji.