AI Trend Research

← AI glossary

AI glossary · Beginner · Agents

AI agent

An AI agent is an AI that, given a goal, makes its own plan and uses tools to carry out multi-step tasks directly. It goes one step beyond a chatbot that only answers, to an AI that "gets work done."

About this term
Level
Beginner · Core ideas for first-time readers
Category
Basics
Fields
Agents
Added
2026-10-07
In plain words It is like an assistant who, when told "get my business trip ready," handles everything from searching for flights to booking a hotel and organizing the schedule. However, it can still make mistakes along the way, so it is a good idea for a person to check its work.

How it works

An AI agent usually uses an LLM as its "brain" and connects tools such as search, calculation, file editing and web browser control. It splits the goal into small steps and repeats a cycle of trying a tool, checking the result and deciding the next action. There are also "computer use" agents that look at the screen and operate the mouse and keyboard directly.

The agent's work loopGoal → LLM (brain); LLM (brain) → Run tool; Run tool → Check result; Check result → LLM (brain)(Repeat); LLM (brain) → Goal achieved(When done)Goal'Research this andsummarize it'LLM (brain)Plan · decide nextactionRun toolSearch · calculation ·files · browserCheck resultDid it work? Whatnext?Goal achievedReport resultRepeatWhen done
The agent's work loop The LLM decides the next action, checks the result of using a tool, and decides the next action again. This loop repeats until the goal is achieved.

Why it matters

Agents let AI go beyond providing information to actually handling work on someone's behalf. Their use is growing quickly in areas such as coding, customer service, document processing and research. That is why many companies see agents as the center of the next generation of AI services.

Things to know

Because agents take real actions, their mistakes have a larger impact. Things like wrong payments, deleted files or unintended posts can happen. So it is important to limit permissions and design agents to get human approval before important actions. Efforts to establish security evaluation criteria also come from this concern.

Example

As in a news item on this site, the Financial Security Institute, a Korean financial security agency, was reported to have established security evaluation criteria for AI agents used in the financial sector. This is a case where, as agents become able to handle customer information and financial transactions, inspection criteria became necessary to use them safely.

Where it came up on this site

Related terms

LLM MCP Prompt Alignment Reasoning model Multimodal