Dictionary.com shortlisted agentic for its 2025 word of the year. Not surprising, given that everybody is selling agents these days. Salesforce says you can hire them the way you hire employees. Microsoft imagines companies that are “human-led and agent-operated.” Sam Altman promised that AI agents would join the workforce.
All of which got Danah Henriksen and me wondering, in the latest installment of our Rethinking Creativity and Technology in Education column in TechTrends, what an agent actually is. And what the relationship is between these three seemingly similar words: agent, agency and agentic.
Agent is a hard-working word. Consider the company it keeps. A reagent in a beaker. Sarin, a nerve agent. Agent Orange over Vietnam. The real estate agent who sells your house, the literary* agent* who sells your book. James Bond, secret* agent.* Aldrich Ames, double* agent.* A point guard at the end of his contract, free agent. A travel agent booking your flight to Denver, and now the software that books it for you.
The key to understanding the word is that in every one of these examples the agent acts on somebody else’s behalf. The nerve agent does not decide to poison you. The secret agent follows orders. Even the double agent, the most autonomous sounding of the bunch, is defined by serving two masters rather than none. And most importantly, none of them has agency.
Agency, in contrast, is not a role but a right, the standing to act for yourself and have your choices count. Its history is a fraught one, since agency never comes free. It can be taken away, and it has often had to be fought for, in courtrooms and classrooms, by people who had been denied it. We describe this history in greater detail in the paper.
Which brings us to agentic. The word was popularized by Stanley Milgram, he of the electric shock experiments. He called the condition in which people hand over their judgment to others, becoming instruments of another’s authority, an agentic state. Agentic, in his use of the term, means unquestioning obedience.
In our paper we argue that understanding these distinctions is critical, because though they may sound similar, they have very different implications for us as educators.
So when the pitch decks say agentic, we should be careful not to hear agency. Milgram’s meaning, in this context, may be the more accurate one. An AI system that is warmer than any teacher, more patient than any tutor, and never disagrees with you is rather good at producing the state Milgram described.
This is not a small matter for educators, because agency has been our word for a century. Dewey argued that education has to connect with the child’s own initiative or it becomes mere pressure from without. Freire insisted that students help set the purposes of their own learning. It is the vocabulary we have used to push back against scripted curricula and teacher-proofed instruction, and it is why a fourteen-year-old sits at her own IEP meeting, so that her goals are negotiated with her rather than for her.
Which is exactly what is at risk when the word gets borrowed.
We must also remember that every agent works for a principal. And who the principal is may not as clear we think it is. Educational agentic AI systems may appear to be addressed to the learner while districts, vendors and platforms decide upstream what counts as progress. That makes them double agents, seemingly working for the learner while actually “taking orders” from someone outside of the learning interaction. Which is why it is worth asking of any agent that turns up in a classroom whom it is really working for.
Three words, with significant surface similarity but with very different meanings, histories and implications. As Mark Twain said:
Citation and link to paper (and some of my related writing) given below.
Mishra, P., & Henriksen, D. (2026).
[Agentic AI in education: Whose agent? Whose agency?]TechTrends, 70(4), 799–805.[https://doi.org/10.1007/s11528-026-01213-1]
Previous posts in this thread
Control vs. Agency: Exploring the History of AI in Education(March 2025)From Symbols to Statistics: The Parallel Histories of Machine and Human Learning(May 2025)The Curiosity Paradox: How Sycophantic GenAI May Undermine Learning(January 2026)The Mirror and the Black Box: AI Metaphors and What They Mean for Learning(May 2026)