What is an agent?

We think of an agent as composed of three things: its data, its wisdom and where you can talk to it.

What is its data?

Its data is the information it has access to when you ask it a question. That could be data we provide, documents you give it, websites or information from your own systems. It gives the agent something to work with.

A company research agent might use company records and financial accounts. An agent helping your team might use your handbooks, product information and internal documents. The data you connect should fit the job you want it to do. You can read more about this in Add knowledge, data & tools.

What is its wisdom?

Its wisdom is the guidance you give it about how to do its job. What should it pay attention to? How should it use the information it finds? What should it ask you before going any further? This is where you explain how you want it to work.

For example, you might tell a research agent to compare companies using the same criteria, show where its information came from and say when something is missing. You can refine that guidance as you use the agent and see what needs explaining. Its owner can edit its wisdom in Manage.

Where can you talk to it?

You can open an agent on the web and start a conversation. Depending on how it is connected, you can also talk to it in a messaging app, send it an email or use it through another AI assistant. The idea is that you can reach it in the places you already work.

For example, your team might use an agent in Microsoft Teams, while you prefer to open it in your browser. Look at the connections offered by that agent to see what is available. The guides cover messaging apps, email and connecting through another AI assistant.