What Is an Agent?
A chatbot answers your questions. An agent does the work. That is the difference between the two. This drastically changes what AI can do inside your business.
Most people first met AI as a chatbot. You type a question, it writes an answer, and then it waits for you. Ask one to write a reminder letter for customers with overdue invoices and you will get a good letter. You still have to look up who is overdue, send each letter and follow up. The chatbot wrote something. You did the work.
An agent starts with a goal instead of a question. You tell it what you want done. It works out the steps, opens the files and programs it needs and checks its own results. It keeps going until the job is finished or it reaches something it is not allowed to do.¹
Claude, an AI made by a company called Anthropic, is a useful example because it can be either one. Give it a task and it can read and change files in folders on your computer, work in your web browser and connect to programs such as your email and calendar. It can repeat a task every day or every week without being asked.² On some subscription plans the work runs on Anthropic’s computers instead of your own, so it keeps going after you close your laptop. You can check on it from your phone or another computer.³ The details differ by plan and are changing quickly.
The uses are ordinary, and that is the point. Anthropic’s own examples include turning a folder of receipts into an expense report, pulling the action items out of meeting notes, and building a weekly report from spreadsheets.² Return to the overdue invoices. An agent that can open your list of unpaid accounts and your email could then find who is overdue and write each reminder for you to approve.
What matters most is how the agent is set up, because it can do only what you allow. You choose which folders it can open and which programs it can reach. You decide whether it must stop and ask before each action or may keep working. It is much like bringing on a new assistant. You decide which keys they get, what they can sign and when they must check with you.
This ability brings a new kind of risk. A chatbot’s wrong answer stays on the screen until you act on it. An agent’s mistake may already have sent the wrong email or changed the wrong record.¹ Anthropic tells its customers that they remain responsible for what Claude does for them.⁴ I covered that responsibility in Take Ownership of AI’s Mistakes. Start with low-risk work, have the agent ask before it acts, and stay close to anything that involves money or messages sent in your name. Keeping agents secure is a subject for a later article.
Here is a simple test for finding a place to start. Think of a task that follows the same steps each time and uses the same files and programs. Could you hand it to a new assistant with one page of written instructions? If so, an agent may be able to do it. A chatbot will tell you how the task could be done. An agent will do it, within the limits you set.
Sources
¹ Jack Pittas, “Agentic AI vs. generative AI: Key differences and use cases,” Zapier, August 3, 2026.
² Anthropic, “Get started with Claude Cowork” and “Schedule recurring tasks in Claude Cowork,” Claude Help Center, viewed October 7, 2026.
³ Anthropic, “Use Claude Cowork on web, desktop, and mobile,” Claude Help Center, viewed October 7, 2026.
⁴ Anthropic, “Use Claude Cowork safely,” Claude Help Center, viewed October 7, 2026.