Agentic AI

What it means

Most people first met AI as a chatbot. You ask, it answers, and you decide what to do next. Agentic AI is different because it is given a goal and works on it: it plans steps, uses tools such as email or company data, and carries the work forward with limited supervision. IBM's definition puts it this way: "Agentic AI is an artificial intelligence system that can accomplish a specific goal with limited supervision."¹

An everyday example: a chatbot can draft a reply to a customer. An agentic system can read the complaint, look up the order, issue the refund, write the reply, and log what it did. The change is about who acts. With agentic AI, the software does the work, and people decide how far it may go.

Where the word comes from

The adjective became common in AI in 2024. Andrew Ng, a well-known AI researcher and educator, explained why he chose it in a June 2024 letter. He had seen arguments about which systems really counted as "agents," and he wrote: "Rather than having to choose whether or not something is an agent in a binary way, it would be more useful to think of systems as being agent-like to different degrees." He added: "Unlike the noun 'agent,' the adjective 'agentic' allows us to contemplate such systems and include all of them in this growing movement."²

OpenAI's policy researchers were describing the same idea at about the same time. In an April 2024 talk on OpenAI's paper Practices for Governing Agentic AI Systems, Yonadav Shavit defined "agenticness" as "the degree to which an AI system can adaptively achieve complex goals," which he said matters most "with limited direct human supervision."³

The word's meaning has since shifted. In 2024 Ng noted that "mainly only technical people use the word 'agentic' (for now!)," and that an article about agentic workflows was "less likely to be marketing fluff."² A year later it had become a sales term. We have not traced the word's older uses outside AI to a source we could open, so we leave that history out.

How others define it

• IBM: "Agentic AI is an artificial intelligence system that can accomplish a specific goal with limited supervision."¹
• OpenAI (Shavit): agenticness is "the degree to which an AI system can adaptively achieve complex goals," most important "with limited direct human supervision."³
• Andrew Ng: "there are different degrees to which systems can be agentic."²
• NIST (the U.S. National Institute of Standards and Technology) does not use the word "agentic" in its February 2026 announcement. It describes "AI agents capable of autonomous actions" that "can function securely on behalf of its users."⁴

Where they agree and differ

They agree on two points: agentic AI pursues goals rather than just answering, and it does so with less human supervision. They differ on whether "agentic" is a category or a matter of degree. IBM describes a kind of system that "consists of AI agents."¹ Ng and OpenAI treat it as a matter of degree, so one system can be a little agentic and another very agentic.²³ NIST talks about agents acting "on behalf of its users," which puts the question of whom the agent works for at the center.⁴

The strongest opposing view is that the word is now mostly marketing. Reuters, reporting a Gartner forecast, wrote that "More than 40% of agentic artificial intelligence projects will be canceled by the end of 2027." Gartner warned of "agent washing," which it called "the rebranding of products such as AI assistants and chatbots without significant agentic capabilities."⁵ That warning is fair, and it is a reason to ask what a product actually does, not to drop the idea. We are fairly confident in the forecast's wording because a major wire service reported it. It is still a prediction, not a measured result.

In the news

On October 8, 2026, Google said it was bringing Gemini into "the agentic age" with "a unified agent that can not only answer questions but also get things done on the user's behalf," starting with businesses.⁶ One detail matters for leaders: "As it takes action, it writes an audit trail attributed to the agent, instead of a person."⁶ The report does not describe a required human sign-off before the agent acts.⁶ When an agent's actions are recorded under its own name, someone still has to answer for them. Organizations have to decide who that person is.

What it means for your organization

When a vendor or a colleague says "agentic," ask what the system can do without a person: which tools it can use, what it can spend, and what it can change. Because agentic is a matter of degree, you can choose the degree. Start narrow, keep a named person responsible for each agent (see [a person in the middle](/definitions/a-person-in-the-middle) and [principal](/definitions/principal)), and widen its freedom as its record earns trust. Mike Lee's book calls this approach "Frame, Then Free" (chapter 7).⁷

Question for your team: For each AI tool we call agentic, who is the one person who answers for what it does?