AI Agents Need Management, Not Just Prompts, Companies Learn

The shift from AI user to AI manager

Most enterprise AI training focuses on prompt engineering—teaching employees how to extract better outputs from generative systems. That approach worked when AI simply responded to instructions. But as AI agents become autonomous, pursuing goals and triggering actions without constant human input, the employee's role fundamentally changes from user to manager.

According to a commentary by Keith Ferrazzi and Wendy Smith published in Fortune, this represents one of the most critical capability gaps in enterprise AI adoption. Companies are teaching people how to get more from AI, but not how to lead it.

Why it matters

The risk isn't obvious failure—it's that agents will appear competent and confident enough that employees relax their judgment precisely when sharper oversight is needed. A 2025 MIT Sloan Management Review and BCG study found 76% of executives view agentic AI as a coworker rather than a tool, yet most organizations lack frameworks for what that relationship actually requires.

Five mental models for agent relationships

Ferrazzi and Wendy Smith, head of research at Ferrazzi Greenlight, identify five distinct ways employees relate to AI agents, each requiring different management approaches:

Tool: The human operates the system for bounded, repeatable tasks—summarizing transcripts or reformatting data—and remains fully responsible for results.

Intern: The agent receives context and close inspection, with gradually expanding responsibilities. Like Wharton professor Ethan Mollick's "AI intern" concept, it can draft documents or conduct research, but humans must review work and provide judgment the agent lacks.

Service provider: The human defines outcomes, scope, and constraints while the agent executes within those parameters. Hala Jalwan, CEO of procurement startup Rivio.ai, notes this makes the human role "more managerial, not less important."

Teammate: The agent participates in ongoing collaboration, developing ideas and coordinating work. BNY's approach illustrates this: the bank onboards "digital employees" with the same governance rigor as enterprise systems, while accountability always remains with people, according to CIO Leigh-Ann Russell.

Expert: The agent provides specialized analysis or recommendations beyond human capabilities. Procter & Gamble research showed AI helped employees incorporate expertise outside their specialization, though humans still assessed whether recommendations fit business context.

The overtrust problem

Boston University research by Emma Wiles found people caught 18% fewer errors when work came from an "AI employee" versus a "chatbot." Rather than abandoning teammate language, organizations should take it more seriously—implementing clear standards, escalation protocols, and ownership structures.

Gianpaolo Barozzi, Cisco's 3P CTO, explains that agentic AI "requires new ways of setting boundaries, calibrating trust, and maintaining human accountability."

The same agent might function as an expert analyzing datasets, a service provider executing procurement workflows, a teammate solving problems collaboratively, and an intern navigating ambiguous situations—all depending on task, context, and demonstrated reliability.

Building human-agent fluency

Organizations need what Ferrazzi and Smith call "human-agent fluency": the ability to recognize which relationship the work requires and manage accordingly. This differs fundamentally from prompt engineering. Instead of asking "How do I get a better answer?" employees must ask "What role should this agent play, and what does that require of me?"

Companies should make expectations explicit for each model—defining what agents may decide independently, how work will be reviewed, and what would justify changing the relationship. Employees must learn to recognize when they're granting too much or too little trust.

Across every model, human accountability for outcomes remains constant. A workforce isn't ready simply because employees use AI frequently—it's ready when people can identify the appropriate relationship, manage it effectively, and recalibrate as tasks and capabilities evolve.

These details were first reported by Fortune in a commentary piece by Keith Ferrazzi and Wendy Smith.