A chatbot answers. An agent acts. The gap between those two verbs decides what you should pay for it, and most ten-person companies are being sold the wrong one.
Chatbot vs AI agent: the short answer
A chatbot replies to messages from a fixed script or a knowledge base; an AI agent can look things up, use tools, take actions in your systems and remember context across a conversation. If the job is answering the same forty questions politely, buy a chatbot. If the job is moving a customer from enquiry to booked appointment, quoted price or updated record without a human copying data between windows, you need an agent, and you should budget for it as software rather than as a widget.
What a chatbot actually does
Most chatbots sold to small businesses are one of two things: a decision tree with buttons, or a language model pointed at a folder of documents. Both are retrieval. The visitor asks, the system finds the closest matching answer, and the conversation ends there. Opening hours, delivery zones, return policy and halal certification status are all questions a retrieval bot answers at two in the morning. The limit is that a chatbot cannot do anything with the answer. It cannot check whether Thursday at three is free, it cannot create the customer in your CRM, and it does not know that the same person asked about the same product yesterday.
The tell is what happens when the question is slightly off script: a chatbot guesses or hands over, and a person still has to finish every conversation that mattered.
What makes an AI agent different
An agent adds four things on top of retrieval. Tools: it can call your calendar, your CRM, your invoicing system or a pricing sheet through an API. Actions: it can write, not only read, so a booking gets created and a lead gets logged. Memory: it carries context across turns and, when designed for it, across sessions, so the returning customer does not start from zero. Reasoning: it decides which of those tools to use, in what order, and when to stop and ask a person. An agent that never escalates is not brave, it is dangerous.
Gartner has warned that many vendors are engaged in what it calls agent washing, rebranding existing assistants, RPA and chatbots as agents without substantial agentic capability, and estimated in June 2025 that only around 130 of the thousands of agentic AI vendors are real. The same release predicted that over 40 percent of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls, according to Gartner. Both numbers say the same thing: know which of the two you are buying before you sign.
Where the money goes
A chatbot costs what its subscription costs: a hosted platform, a monthly fee, a few days of setup to load your documents, and light maintenance when prices change. The build is cheap because there is almost nothing to build.
An agent has two cost lines that behave differently. The build cost is integration work: connecting the model to your calendar, CRM, payment or e-invoice system, writing the rules for what it may and may not do, and testing the edge cases where it should hand over. The running cost is model usage plus the platform that hosts it, and it scales with conversation volume and with how much reasoning each conversation needs. In our experience the build is where the budget sits and where payback is decided, because an agent that acts on ten conversations a day replaces part of a role, not part of a FAQ page.
Chatbot vs AI agent for a ten-person company: how to decide
Start with the conversation that costs you the most, not the one that happens most often. For most small businesses that is the enquiry that arrives while everyone is busy and quietly goes to whoever answered first. Write down what a perfect response to that enquiry involves. If it is a paragraph of information, a chatbot is enough. If it is a check against a system, a decision and a record written somewhere, it is agent work.
Then count the tools. If the answer lives in one document, buy retrieval. If it lives in three systems a person reconciles by hand, you are paying that person to be the integration layer, and an agent is cheaper within months. Finally, define the hand-over. The best small-company agents do one bounded job completely, book, quote, qualify or update, and pass everything else to a named human with the full context attached. That boundary is what keeps the project out of Gartner's cancelled 40 percent.
Five questions that separate the two before you pay
- What can it write, not just read? Ask for a demo that ends with a record created in your CRM or a slot booked in your calendar. If it ends at a nicely worded reply, it is a chatbot regardless of the label on the invoice.
- What happens when it is unsure? A real agent has an explicit escalation path that hands the conversation history to a person. Test it with a question that is half in scope.
- Does it remember the customer? Ask whether the second conversation with the same WhatsApp number starts with context or from zero.
- Which systems does it need access to, and who controls that access? Every tool connection is both a capability and a risk: you want scoped credentials, an audit log of actions, and a kill switch you can reach without a developer.
- What is the build cost versus the monthly cost, separately? Agents should be quoted as a build plus a run cost, and the run cost should be tied to volume you can forecast.
Chatbot vs AI agent: common questions
- Is an AI agent just a smarter chatbot? — No. A chatbot retrieves and replies; an agent retrieves, reasons, uses tools, takes actions in your systems and keeps memory across a conversation. The practical test is whether it can complete a task, such as booking an appointment or creating a CRM record, without a person finishing the job.
- Does a small business need an AI agent or is a chatbot enough? — If your most valuable conversations end in a paragraph of information, a chatbot is enough. If they end in a booking, a quote, a qualified lead or an updated record that a person currently types by hand, an agent pays for itself faster because it replaces work, not only answers.
- How much more does an AI agent cost than a chatbot? — A chatbot is mostly a monthly subscription with a few days of setup. An agent has a one-time build cost for integrating your calendar, CRM or invoicing tools and writing its rules, plus a running cost that scales with conversation volume. The build is the larger line and the one that determines payback.
- What is agent washing? — Agent washing is the term Gartner uses for vendors rebranding chatbots, assistants or RPA tools as AI agents without real agentic capability. Gartner estimated in June 2025 that only around 130 of the thousands of agentic AI vendors are genuine, so ask for a demo that ends in an action, not a reply.
Keep reading
Build something
worth trusting.
Tell us what's slowing your business down. We'll show you the system that fixes it — and how fast.