このページはまだ翻訳されていないため、英語版を表示しています。
基礎知識

What is an AI agent?

Software that can hold a conversation, look things up, and take actions on your systems — deciding for itself which of those to do next.

別の呼び方agentAI assistantautonomous agentagentic AIAI agents

A chatbot answers. An agent decides.

Both take a question and produce a reply, so from the outside they can look the same. The difference is what happens in between. A chatbot maps your message to a response — from a script, or from whatever a language model produces. An agent is given a goal, a set of tools, and a boundary, and it works out the steps itself: look this up, ask a clarifying question, call that system, then answer.

The three things an agent has that a chatbot doesn't

Tools it can actually call. Not "here is a link to our booking page" but placing the booking. An agent is connected to real systems and can act in them, which is why the interesting question about any agent is what it is permitted to touch.

A loop. It can take a step, look at the result, and decide the next step. If the first search returns nothing useful it can search differently. A chatbot has one shot.

A memory of the relationship. Not just the last few messages, but what it learned about this person the last time they spoke — and the time before.

"Agentic AI" is the same thing

The industry settled on agentic AI for this category — software that plans, uses tools and acts rather than only replying. If you arrived here searching for that, you are in the right place: the pages in this section describe an agentic system, built for one business rather than as a general assistant.

We tend not to use the phrase in our own writing, because it has been attached to enough products that it no longer narrows anything down. What narrows it down is the list above: tools it can actually call, a loop, and a memory of the relationship.

Where the word gets abused

"Agent" is now attached to almost anything with a text box, so it is worth being blunt about what does not make something an agent: being powered by a large model, having a personality, or being able to answer in seven languages. Those are all features of chat interfaces. If it cannot do anything except talk, it is a chatbot with better prose.

What an agent is made of here

Those three requirements are not abstract on this platform — each is a thing you configure, and the rest of this section is one page per piece:

The agent…Comes from
knows your businesstwo knowledge bases — your vetted material and each customer's own documents
behaves like your businessits soul, under a moderation layer no tenant can disable
follows your proceduresSkills, loaded only when the conversation needs them
reaches your systemstools over MCP, enabled only through a loaded Skill
knows which page you came froma page playbook
collects facts cleanlyask forms it composes on the spot
acts in the gapsscheduled tasks that fire months later, in your customer's time zone
keeps customers aparta space each, enforced in the database

If you are weighing this against a chatbot, the row that decides it is usually the last two.

The practical test: ask what happens after the conversation ends. If the answer is "nothing", you are looking at a chatbot.

中の仕組みも知りたいですか?How the tool loop works
ほかの基礎知識