Does agentic AI fit the tool-or-coworker framework?

MIT Sloan Management Review and Boston Consulting Group's ninth annual global study on AI and business strategy argues that agentic AI "complicates" the categories executives have long used to frame technology: "Tools automate tasks, people make decisions, and strategy determines how the two work together." Because agentic systems "plan, act, and learn on their own," the report finds 76% of respondents to its global executive survey "view agentic AI as more like a coworker than a tool." The finding comes from a spring 2025 survey analyzed from "2,102 respondents representing more than 21 industries and 116 countries," plus 11 executive interviews across financial services, technology, retail, energy, and health care.

The report's mechanism is that this "tool-coworker duality" pulls functional roles in different directions at once: "IT leaders want predictable, scalable systems with clear technical specifications. CFOs need investment models with measurable returns and depreciation schedules. HR executives require... performance management frameworks and supervision protocols." Because a single agentic system can simultaneously substitute and complement, and behave as both capital and labor, MIT SMR and BCG argue it demands "both human resource approaches and asset management techniques" at once — something the traditional substitute-or-complement framework, built on an either/or assumption, cannot accommodate. The report frames this as producing "operational tensions" leaders must manage rather than resolve, naming "scalability versus adaptability" and "experience versus expediency" among them.

This gives a structural reading to findings already in the library. Why are companies deploying agents faster than governance matures? and Why do individual AI gains not translate to business results? both document adoption outrunning the frameworks meant to capture or govern it; MIT SMR and BCG's duality argument supplies one reason why — governance models and EBIT accounting built around a single category (tool or worker) will systematically miss the half of agentic AI that behaves like the other. The duality also bears on Does personal preference shape how engineers use AI tools?, since a company's agency policy is exactly where the tool-or-coworker framing gets decided in practice, and it parallels Can AI boost how teams work together? in arguing that agentic systems exceed the categories built for individual tool use and require new organizational design rather than a better-tuned version of the old one.

The excerpt does not detail how the "coworker" perception was measured beyond the single survey item, nor does it give the other two of the four named operational tensions (only "scalability versus adaptability" and "experience versus expediency" appear in this excerpt). The 76%, and the 35%-adopted-with-44%-planning adoption figures, are MIT SMR and BCG's own survey self-reports, not independently verified organizational outcomes, and BCG is a management consultancy with a commercial interest in defining a new strategic-framework problem it is positioned to advise clients on. What the excerpt supports is that surveyed executives describe agentic AI in coworker terms and that organizations are adopting it faster, at a comparable stage, than they adopted traditional or generative AI — not that the tool-coworker duality is the only workable lens for the resulting management tensions.