The one-line version
A chatbot answers questions. An agent gets things done.
Ask a chatbot "where is my order?" and it tells you how to find your tracking number. Ask an agent the same thing and it looks up your order, sees the parcel is stuck at a hub, drafts the apology, and offers a replacement — because it has access to systems and permission to act, not just knowledge to recite.
What a chatbot is
A modern chatbot is a language model wrapped around a knowledge source — your FAQs, policies, product docs. It can converse naturally, retrieve the right passage, and summarise it well. That's genuinely useful: a good chatbot deflects the repetitive half of a support queue and never sleeps.
But its limits are structural. It can only tell. The moment a request needs information from a live system ("check my account") or an action ("change my delivery address"), a chatbot has one move: hand you off to a human, or worse, guess.
What an agent is
An agent is a language model with three additions:
- Tools — connections to real systems: your order database, CRM, calendar, payment provider.
- A loop — it can take a step, look at the result, and decide the next step, several times over, until the task is done.
- Guardrails — defined limits on what it may touch, what needs human sign-off, and when it must stop and escalate.
That third item is not optional garnish. An agent without guardrails is a liability with an API key. The engineering effort in a production agent goes mostly into tools and guardrails, not the conversation.
Side by side
The same customer message, two systems:
- "I was charged twice." Chatbot: explains the refund policy, opens a ticket. Agent: checks the payment records, confirms the duplicate, prepares the refund for one-click human approval, replies with a timeline.
- "Reschedule my appointment." Chatbot: links to the booking page. Agent: checks the calendar, offers three slots, books the one you pick, sends the confirmation.
- "What's your warranty?" Both answer it fine — and the chatbot does it cheaper. Not every question needs an agent.
Which one do you need?
A chatbot is enough when the job is answering: your questions are repetitive, the answers live in documents, and nothing needs to be looked up per-customer or changed in a system. It's cheaper to build, cheaper to run, and easier to trust.
You need an agent when the job is doing: requests routinely require live data or actions, the volume justifies the build, and the actions are checkable or reviewable. Most businesses we work with end up with a sensible hybrid — a chatbot front door, an agent behind it for the requests that need hands.
The wrong answer is buying an "agent" because the word is newer. Match the machine to the job.
Not sure which one your business needs?
Describe the job you want done. We'll tell you plainly whether it's a chatbot, an agent, or a script — and build whichever it honestly is.