A founder used AI every day and still spent fifteen minutes cleaning up after every answer.
She drafted client updates with AI each morning. The draft arrived in seconds. Then came her part: copy it out, paste it into the email, fix the formatting, file a copy in the client folder, hit send. Five manual steps, every time.
She listed them one day and realised the AI was doing the writing while she was doing everything else.
So she set up an agent. It drafts the update, formats it, and files it in the client folder. She reviews and sends. Fifteen minutes became two.
The plain version
The word agent gets used loosely, so here it is without the jargon.
Ask someone to look up a phone number and read it back to you. That’s a chatbot.
Ask them to call the number, book an appointment for Tuesday at 2pm, and confirm it by email. That’s an agent.
Same knowledge in both cases. Different job. A chatbot answers questions. An agent takes steps (opens in a new tab).
How to know which you need
You’ve been using chatbots for most of this series. Ask a question, get an answer. Paste a document, get a summary. That works fine when the answer is the product.
The signal that you need more is what happens afterwards. You copy the summary into an email. You paste the draft into your project tool. You type the numbers into a spreadsheet.
Steps you repeat after every answer are the case for an agent. When the answer is the product, a chatbot is enough. When the answer has somewhere to go, an agent does the going.
One safety note before you connect anything: know exactly what data an agent can reach (opens in a new tab). An agent that reads your inbox shouldn’t also reach your financial systems unless it genuinely needs both. Set the boundaries before the first connection, not after the first incident.