26 Sept 2026 · @Gerd Jährling
AI agents already do knowledge work. Whether they are knowledge workers is a different, and more strategic, question.
Peter Drucker introduced the term “knowledge worker” in Landmarks of Tomorrow (1959). He described people who create value by applying theoretical and analytical knowledge rather than manual skill, and predicted they would become the most valuable asset of the 21st-century organisation.
Sixty years later, enterprises are deploying AI agents that analyse, draft, plan, decide and act. In Multi-Agent systems they even hold defined roles, much like members of a team. So the question is no longer academic: if agents do the work of knowledge workers, should organisations treat them as such?
As a researcher working between technology and business, I find Drucker a useful inspiration, precisely because he was never interested in technology for its own sake. He cared about productivity, responsibility and what makes an organisation effective.
Drucker’s test, applied to agents
In “Knowledge-Worker Productivity: The Biggest Challenge” (1999), Drucker named six factors that set knowledge work apart from manual work. They can make a practical test for agents.
| Drucker’s factor | Do today’s agents meet it? |
|---|---|
| Asking “what is the task?” | Rarely. Agents decompose tasks well, but the goal is still set for them. |
| Responsibility for their own productivity, with autonomy | Partly. Agents act autonomously within bounds, but accountability stays with humans. |
| Continuous innovation as part of the job | Partly. Agents can propose new approaches, but seldom question the process they sit in. |
| Continuous learning and teaching | Partly. Agents adapt through memory and feedback, and can document for others, but do not learn the way people build careers. |
| Quality, not just quantity | Yes, and it is measurable, often better than for humans. |
| Treated as an asset, not a cost | This is a management choice, not a property of the agent. |
Drucker made one more observation that matters here: knowledge workers own their means of production, because the knowledge sits in their heads and leaves with them. Agents are the reverse. Their capability belongs to the firm, or to the vendor behind the model. That single difference changes the economics of the whole question.
My position: knowledge co-workers, not knowledge workers
Measured against Drucker, today’s agents are not knowledge workers in the full sense. They perform knowledge work, but they do not define their own task, own their knowledge or carry responsibility for their contribution.
Yet treating them as mere tools misses what is happening. In Multi-Agent systems, agents hold roles, collaborate and make bounded decisions alongside people. In my research on the cognitive transformation of the enterprise, I therefore argue for widening the classic view of a firm’s human resources to include AI agents as knowledge co-workers: contributors whose value depends on the people and routines around them.
That framing has a strategic payoff. Drucker’s knowledge workers were valuable but mobile; they could walk out of the door. Agents grounded in a firm’s own knowledge, data and processes stay. If organisations situate them well, agents may become one of the few knowledge assets a competitor cannot simply hire away. If they deploy them generically, they own nothing a competitor could not buy tomorrow.
So the most Druckerian question is not “are agents knowledge workers?” It is “what is the task?”, and whether the enterprise, not the agent, is ready to answer it.
References
- Drucker, P. F. (1959). Landmarks of Tomorrow. Harper & Row.
- Drucker, P. F. (1999). Knowledge-worker productivity: The biggest challenge. California Management Review, 41(2), 79–94. Link
- Kemp, A. (2024). Competitive advantage through artificial intelligence: Toward a theory of situated AI. Academy of Management Review.
No responses yet