What is agentic AI?

How it works

Definitions vary. The U.S. National Institute of Standards and Technology (NIST) issued a January 2026 request for information. It said AI agent systems can plan and take autonomous actions that affect real-world systems or environments. By its description, such a system pairs at least one generative AI model with scaffolding and can be deployed with little or no human oversight. Scaffolding is the surrounding software that gives the model tools to act with. The notice treats AI agents and agentic AI as other names for the same thing.

Developers draw finer lines. In a December 2024 engineering guide, Anthropic grouped two designs under the label agentic systems. In workflows, models and tools follow paths fixed in code. In agents, the model directs its own steps and tool use. The company added that such systems often trade speed and cost for better task performance.

Why it matters

NIST said in February 2026 that AI agents could by then run on their own for hours and handle jobs such as:

• writing and debugging code
• managing email and calendars
• shopping

The January request for information says such systems may be open to hijacking, backdoor attacks and other exploits. It gives indirect prompt injection as an example of an attack that arrives through adversarial data a model handles.

Where things stand in 2026

OSWorld tests agents on computer tasks across operating systems. Stanford University's 2026 AI Index says scores on it climbed from about 12% to 66.3%. It says agents still fail roughly one in three attempts on structured benchmarks.

NIST's Center for AI Standards and Innovation launched an AI Agent Standards Initiative on February 17, 2026. It centers on industry-led agent standards, community-led open source protocols, and research on agent security and identity. On December 9, 2025, the Linux Foundation announced the Agentic AI Foundation. It called this a neutral home for open source agent projects including the Model Context Protocol.

The research firm Gartner has questioned some commercial claims. In June 2025 it predicted that more than 40% of agentic AI projects would be canceled by the end of 2027. It cited rising costs, unclear business value or inadequate risk controls. Gartner also described agent washing, the rebranding of existing products such as assistants and chatbots as agents. It estimated that only about 130 of the thousands of agentic AI vendors were genuine.