An AI agent pursues goals and takes actions, not just answers questions

An AI agent pursues goals and takes actions, not just answers questions
Image: Intellect (generated)

In short: Most chatbots answer a question and stop. An AI agent is built to work toward a goal across many steps, using tools and making decisions along the way. Agentic AI is now being applied to tasks like travel booking, web browsing and software development, and the field is still working out what the label even means.

An intelligent agent, in the general sense, is anything that perceives its surroundings and acts to reach goals, from a basic thermostat to a person. In artificial intelligence, AI agents, also called agentic AI, extend this idea by actively chasing goals, making choices and carrying out actions across long stretches of time. What steers them is an objective function that encodes what they are meant to achieve. A reinforcement learning agent works from a reward function, an evolutionary algorithm from a fitness function, and a self-driving car's objective has to weigh several competing aims at once, such as safety, speed and rider comfort.

Researchers describe agents in layers of growing capability. Simple reflex agents respond only to what they currently sense, while model-based agents keep an internal picture of parts of the world they cannot observe. Goal-based agents pick actions that lead to a desired state, and utility-based agents rank outcomes by how desirable they are. Many modern LLM-driven agents are described as goal-based, with a language model often directing their sequence of steps. They commonly add memory, planning logic and tool interfaces, coordinated by orchestration software.

The term has no settled definition. Commentators note that there is no universally agreed meaning, and the word agentic only became common in 2024, partly through the influence of researcher Andrew Ng. Earlier products such as Siri and Alexa were sometimes called agents, though they lacked the broad reasoning of later language-model systems. The Financial Times compared agent autonomy to the SAE levels used for self-driving cars, placing most current applications at level 2 or 3.

Practical uses are spreading. Commerce protocols such as the Agentic Commerce Protocol, developed by OpenAI and Stripe, define how agents can search products, manage carts and start checkout. Agentic browsers can navigate pages and fill in forms on a user's behalf, though AI browsers have been noted to be vulnerable to prompt injection attacks. Governance is emerging too, with Singapore's IMDA publishing a framework for agentic AI in January 2026.

Opens the Intellect+ feed with this card first. Swipe up for the next one.

Sources

To cite this card, link to https://intellectplus.co.uk/app/technology-agentic-ai.

How we source, check and correct cards: Editorial standards.

General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.

Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?

Related

Finding something worth knowing…