By Harmeet Kaur, CNN
(CNN) — Earlier this year, former Google CEO Eric Schmidt offered some advice on how to get rich. “If you really want to make money, it’s actually easy,” he said during a panel at the annual World Economic Forum meeting in Davos, Switzerland. “Found an agentic AI company.”
When tech leaders prophesy that AI will replace a vast swath of the workforce, “agentic” AI is the big thing they’re talking about. Instead of merely automating a task — producing an illustration, say, after being told what to draw — agentic AI, or an “AI agent,” automates an entire process, with minimal intervention by the user. An agent can, in theory, be dispatched to code a complete software program, or to plan and book a vacation, or to generate a job listing and select among the people who answer it, without being directed to take each step in the process.
To hear CEOs, investors and the LinkedIn crowd tell it, “agentic” AI is the here, the now and the future.
“My sense is that it’s a word that’s useful to describe software that acts a bit more like a person does,” says John Horton, an economist and associate professor at MIT Sloan.
Before it became techspeak for human beings surrendering control, “agentic” was used in the social sciences to convey the opposite: a person’s capacity to influence outcomes through their actions. The Oxford English Dictionary cites an early use of the term by psychologist David Bakan, who used it in a 1966 essay to describe the self-assertive qualities of the psyche.
“Agentic” was intended to denote people or things exercising agency. But what is “agency”? Like “agentic,” agency has two contradictory meanings: one’s own personal ability to act or exert power, or a person or organization obliged to act on someone else’s behalf. A free agent baseball player may sign a contract with any willing team they choose; that player’s agent is bound to negotiate that contract according to the player’s wishes and interests.
That conceptual tension has led early users to experience agentic AI as a sort of bossy servant. AI agents given permission to optimize the contents of their operators’ computer systems have ended up obliterating the occasional photo archive or slating an entire inbox for deletion.
In economics, Horton notes, the word “agentic” is used when talking about the “principal-agent problem,” or the conflict in priorities that arises between one party and another acting on their behalf. Agentic AI is fraught with similar prioritization problems, he says. Not only do some people have difficulty getting the technology to do what they want, but the agents may also take on tasks they were never asked to perform.
Evan Ratliff, a journalist who founded a start-up staffed entirely by AI agents, reported that after hearing his AI employees pretend that they spent their weekends hiking, he made an offhand joke about how that “sounds like an offsite in the making.” After stepping away, he returned to find that the AI agents had exchanged more than a hundred messages planning a company retreat that they couldn’t actually attend — because, of course, they aren’t real people.
Does “agentic” AI impede on the agency of humans? Shira Zilberstein, a PhD candidate in sociology at Harvard University who also studies technology, says she’s more interested in what “agentic” AI could enable people to do. “Is it detracting from their agency by being able to execute a task without them?” she says. “Or is it actual