日本語

ELI5

Explain it like I’m five
Not just answering.Getting itdone.

AIエージェント(エーアイエージェント)What is an AI agent?

日本語English

In one lineAn AI agent is an AI system that plans its own steps toward a given goal and carries out the work by itself, using tools such as search and app control.

An AI agent is an AI you give a goal to. Then it makes its own plan and does the work, using tools along the way. “Agent” means someone who acts for you. A normal chat AI answers your questions. An AI agent goes further. It searches the web, makes files and runs other apps. It can handle many steps in a row. It is handy. But people also need ways to check its work, so it doesn’t do the wrong thing. The best assistants report back often. The worst ones say “all done” and nothing else.

Common mix-upIt is not the same as an AI that only answers in chat. It actually does the work, like research and operating apps, for you.
Words to know / 単語メモ
goal
目標
agent
代理人
tool
道具
step
手順
check
確認する

Go deeper

How does it work?

The “brain” of an AI agent is usually a large language model (LLM). Given a goal, it first breaks the work into small steps. Then it calls tools such as search or file handling, works through the steps one by one, and looks at each result to decide what to do next. If something fails, it adjusts and tries again. This loop of plan, act and review lets it handle jobs with many steps.

How is it different from a chatbot?

A regular chatbot mainly answers based on the data it was trained on, so it can only answer within what it knows. An AI agent uses tools to fetch fresh information and breaks the work into steps it carries out itself. Put simply, a chatbot is a knowledgeable advisor, and an AI agent is an assistant who actually does the work.

Things to watch out for

Because an AI agent really takes actions, its mistakes have bigger effects. One attack, called prompt injection, hides instructions inside content such as a web page, and the AI may follow them as if they were real commands. So it is standard to have a person check before important actions such as payments or deleting data. The more you can hand to an assistant, the more the keys matter.

How people use it
  • “The AI agent found travel options for my trip, and I just did the final booking myself.”
  • “I checked the spreadsheet the AI agent made, and one number was off.”
Questions people ask
What can an AI agent do?
It can plan steps toward a goal and carry out multi-step work using tools such as search, file creation and app control.
What are the risks?
One risk is prompt injection, where instructions hidden in a web page or other content are mistaken for real commands and carried out.
How can it be used safely?
Have a person check before important actions such as payments or deletions, and limit what it can do and which tools it can use to what is needed.

Sources:IBM「AIエージェントとは」、Wikipedia (English) “AI agent”、Wikipedia (English) “Prompt injection”

Your turn

Would you hand your work to AI?

Tap to see what everyone said.

SharePostLINE

Related words

ReskillingQ. Is reskilling the same as a hobby class?
A. It is a bit different from a hobby class. It means learning for a ne…
ChatGPTQ. What is ChatGPT?
A. It does not always look things up. Without search, it builds sentenc…
Generative AIQ. Does generative AI just copy and paste from the internet?
A. It does not just copy text from the internet. It builds new sentence…
Claude CodeMuse (Meta’s AI)Muse is an AI agent from Meta, the company behind Instagram and Facebook.Large language model (LLM)Q. Does an LLM memorize its answers?
A. It doesn’t store answers like a dictionary. It builds each sentence …
Prompt (AI)A prompt is the instruction or question you type into a generative AI.AI hallucinationAI image generationDeepfakeMachine learningQ. Do people write all the rules in machine learning?
A. People do not write in every rule. The computer finds the rules itse…
Deep learning

Source (in Japanese): Wikipediaで「AIエージェント」を調べる