Share with your CHRO
Framing AI agents as “coworkers” isn’t just a branding choice, it’s a measurable liability. Boston University researcher Emma Wiles found that managers caught 18% fewer errors in AI-generated work when it was labeled as coming from an “AI employee” rather than a chatbot, and were 44% more likely to escalate questionable outputs upward rather than correct them directly. Nearly a third of the 1,261 managers she studied already work at companies that frame agents as employees, some listing them on org charts.
What this means for your business
Whether your organization is in this trap depends less on your AI strategy than on how your managers talk about AI day to day. Companies that have leaned into the “digital colleague” framing pushed by vendors, including Microsoft, OpenAI, and Anthropic, have essentially run an uncontrolled experiment in accountability diffusion, where the label on a tool changes how much responsibility a human feels for its outputs. If your workforce training treats agents as teammates rather than instruments, you’ve already shifted who your employees think is responsible when something goes wrong.
The accountability erosion Wiles documents has a specific organizational shape: it climbs. When managers stop trusting their own corrections and escalate instead, they’re not exercising appropriate caution, they’re offloading a judgment they were hired to make. That’s the opposite of the efficiency gains that justify AI agent deployments. The irony is sharp: the more human-like the framing, the less human oversight you actually get. Any CHRO whose organization is deploying agents in consequential workflows, performance reviews, hiring screens, benefits adjudication, needs to audit not just what the tool does but what employees believe the tool is.
The Iran school bombing example in the piece is instructive precisely because it’s extreme. Claude got public blame; the actual failure was a cascade of human decisions, each made by someone who probably felt slightly less responsible because a system was involved. The low-stakes version of that dynamic is already running inside enterprise HR departments right now. The leading indicator to watch isn’t error rates in AI outputs; it’s whether your managers can still articulate, clearly, why a given AI recommendation is or isn’t right. If they’ve stopped being able to explain that, the framing has already done its damage.
Based on reporting from AI agents are not your “coworkers”, originally published 2026-06-29 14:00:00.
