The research shows that framing AI agents as “employees” has several unintended downsides without clear upside on adoption.

In a large-scale experiment, when people were encouraged to see AI as a coworker rather than a tool, organizations saw:

• Reduced individual accountability – people were more likely to assume the AI was responsible for outcomes.
• More unnecessary escalation – issues were pushed up the chain instead of being resolved at the right level.
• Lower review quality – humans checked AI output less rigorously.
• Higher role uncertainty – employees became less clear about what they owned versus what the AI owned.

Importantly, these effects came without improving AI adoption. So you take on more risk and confusion, but you don’t get a corresponding increase in usage or value.

The takeaway: AI agents work better when they are positioned as powerful systems that people supervise and are accountable for, not as peers or employees that “share” responsibility.

The research suggests the core question is not whether to deploy agentic AI, but how to design the surrounding system so humans stay clearly in charge.

Instead of putting AI on the org chart as an employee, consider these principles:

• Keep humans clearly accountable: Make it explicit that managers and employees own decisions and outcomes, even when AI is involved.
• Define AI as a supervised system: Position AI agents as tools that can automate steps, propose options, or monitor activity, but always under human oversight.
• Clarify roles and handoffs: Document where AI is used in a workflow, what it produces, and who must review and approve its output.
• Align governance with risk: For higher-risk use cases, require stronger human review and clearer sign-offs.

This approach helps you reimagine workflows around AI while avoiding the accountability gaps and role confusion that come from treating AI as a “colleague.”

The findings point to a practical shift: focus less on giving AI a “seat at the table” and more on reshaping how work gets done around it.

Key changes to consider:

• Workflow redesign: Map where AI can take on repeatable tasks (e.g., drafting, summarizing, monitoring) and where humans must step in to interpret, decide, and approve.
• Role clarity: Update job descriptions and team charters so it’s clear who is responsible for supervising AI outputs and making final calls.
• Governance and oversight: Establish guidelines for when AI can act autonomously, when human review is mandatory, and how issues are escalated.
• Training and expectations: Train employees not just on how to use AI, but on how to supervise it—how to question outputs, spot errors, and document decisions.

By rethinking workflows, roles, and governance in this way, organizations can use increasingly capable AI systems while keeping human accountability and trust at the center.