A chatbot answers questions from a script. An AI agent does the work. It has a job description, real tools, and a brand-trained voice, and it completes tasks end to end (drafting, scheduling, reconciling, following up) with an approval step before anything goes out. A chatbot talks. An agent finishes things.
That distinction sounds like marketing until you've been on the wrong side of it. So let me make it concrete.
The chatbot you already know
You've met the chatbot. It's the little widget in the corner of a website that asks how it can help and then, no matter what you type, routes you to the same three help articles. It's the phone tree that reads you a menu. It's useful in a narrow way, deflecting the easy questions so a human doesn't have to, and it is completely incapable of doing anything on your behalf.
Ask a chatbot to reconcile last month's invoices and it will hand you a link to a page about invoices. Ask it to draft five LinkedIn posts in your voice and schedule them, and it will explain what LinkedIn is. The chatbot isn't broken. It was never built to do the work. It was built to talk about the work.
For a small team, that's the gap that matters. You don't need something that answers questions about your business. You need something that takes things off your plate.
What an agent does that a bot can't
Picture the task you keep meaning to get to and never do, say the weekly customer follow-ups. A chatbot can tell you follow-ups are important. An agent pulls the list of customers who went quiet, drafts a message to each one in your voice, references what they actually bought, queues them up, and waits for you to hit approve. You skim, you fix one, you approve the rest. Fifteen minutes instead of two hours, and the two hours were the ones you never had.
Four things separate the two, and a chatbot has none of them. Give an agent a job description and it owns a specific role instead of a grab bag of “help”: a finance agent reconciles, and it doesn't write your ads. Wire it into your inbox, your calendar, your CRM, whatever the job requires, and it can act on the work rather than just advise you about it. Train it on how your business actually talks, and it stops sounding like a generic helpful-assistant and starts sounding like you. And because nothing it produces goes out until a human says go, that approval step isn't a limitation, it's what makes handing the agent real work safe in the first place. Strip any one of those out and you're back to a chatbot with extra steps.
“But we already have Slack AI / a ticket bot”
This is the one I hear most, and it's fair. If you work in Slack, you've probably got Slack AI, or Claude in a channel, or a bot that files tickets. Those are chatbots too, good ones, but chatbots. They summarize, they answer, they search. They don't return finished, approved work.
The thing a small team actually wants is to brief a task the way you'd brief a colleague (“follow up with everyone who hasn't reordered this month, here's the tone”) and get back completed work with your approval gate on it. Not a summary of the task. The task, done. That's an agent, and most “Slack AI” options aren't that.
Why this matters more for small teams than big ones
A big company can afford to bolt a chatbot onto a support queue and call it AI. They have the headcount to do the actual work anyway; the bot is a garnish.
A small team doesn't have that slack. When it's you and three other people, the difference between a tool that talks about the work and a tool that does the work is the difference between AI as a novelty and AI as a hire. You don't need another thing to check. You need a teammate who shows up, does the job, and hands it back for a quick look before it goes out.
So if you're evaluating anything that calls itself AI, ask one question: at the end of the interaction, do I have finished work, or do I have another answer? That tells you which one you're actually buying.
Matt writes Field Notes, a newsletter on running AI agents inside a real business. Sign up here.