The AI agent vs chatbot difference is about permission, not intelligence. A chatbot answers questions from what it knows or can look up. An agent decides which step is needed and carries it out in another system. So the choice follows the task type rather than the tool.
This is for a founder or team lead holding a proposal and wondering what is actually being ordered. For what an AI agent is in the first place, read what is an AI agent. This article is about the choice.
What separates an AI agent from a chatbot?
The number of things the system gets to decide. A chatbot holds a conversation and answers. An agent, in n8n’s words, is an autonomous assistant with a language model, instructions and capabilities you configure, such as tools, skills and access to a knowledge base [S-038].
| Chatbot | AI agent | |
|---|---|---|
| What it does | gives an answer | takes an action |
| Who sets the order of steps | the builder, in advance | the system, per turn |
| Touches other systems | to read | to read and to change |
| What a mistake looks like | a wrong answer | a wrong action |
| What you have to set up | content and escalation | content, escalation, permissions and review |
The last row carries the decision. A wrong answer is an inconvenience you can correct in the next message. A wrong action in your CRM or stock system is a correction somebody has to undo. This is a question about permission, not a product comparison.
Which one fits your task?
Start from the task, not from the tool. This table is missing from every comparison article on the subject, and it settles the question in one pass.
| Task type | Example | What to build | What goes wrong with the other choice |
|---|---|---|---|
| Informing, fixed answer | opening hours, return policy | chatbot on your own content | an agent adds risk and nothing else |
| Looking something up | where is my order | chatbot with one read connection | without the connection the model invents an answer |
| Doing one thing | booking a slot, raising a ticket | agent with narrow permissions | a chatbot promises what never happens |
| Several steps in a row | drafting a quote from three sources | agent, with a review step | a fixed workflow breaks on every exception |
| Fixed order, no judgement | moving data, assembling a report | a plain workflow, no AI | an agent makes predictable work unpredictable |
| Judgement or exception | complaint, discount, contract question | a person, possibly prepared by AI | any automated answer costs trust here |
Row five is the one most often skipped. A lot of what gets sold as an agent is work with a fixed order and no judgement in it at all. A plain automation is cheaper there, faster and easier to check. That distinction is covered in automation versus AI automation.
When does the system decide for itself?
With an agent, the model decides on each turn whether to call a tool or answer directly. Anthropic describes the default behaviour: it calls a tool when the request maps to that tool’s described capability and the answer is not already in context, and it answers directly for stable knowledge, creative tasks and conversational turns [S-039]. The model decides when a tool is needed [S-017].
That is less open ended than it sounds. What the system can do is fixed by the contract you write: your application specifies which operations are available and what shape their inputs and outputs take, and the model only decides when and how to call them [S-021]. An agent can do nothing you have not set up. The freedom is in the order, not in the permissions.
What does an agent need in order to act?
Four parts, following the n8n documentation: a language model that reasons, instructions describing the role and its limits, tools it can act through, and memory for the running conversation [S-038]. Without tools you have a chatbot with a more expensive name.
Those tools are connections into the systems you already run. Anthropic names exactly that as the case tool use exists for: calling into databases, internal APIs and filesystems, as the bridge between a natural language request and the system that fulfils it [S-019]. An API is the entry point two systems use to exchange data. Since late 2024 there is also an open standard for those connections, MCP [S-001], explained in what is MCP.
When do you need neither?
When the answer comes from what the model already knows. Anthropic lists that as the case where tool use is not needed: summarising, translating and general knowledge questions do not need a tool round trip [S-020]. For that kind of work a plain model in the hands of an employee is enough, with no connection and no agent.
And when the work follows the same order every time, no model belongs in it at all. A workflow with fixed steps is cheaper and more predictable. The answer to “do we need an agent for this” is no more often than the market suggests.
When should a person take over?
At money, at exceptions and at emotion. Three rules to fix in advance: anything involving an amount or a discount outside the standard goes to a person, a question misunderstood twice goes to a person, and a complaint goes straight to a person. Those rules belong in the instructions, not in the builder’s head.
One more thing helps. Have the system say that it is handing over, and to whom. A conversation that stalls is worse than one that transfers.
How long before something like this runs?
Less time than the common assumption. WeAdapt works from discovery call to live in four weeks, including process mapping, building, review and training. That covers a first working application with a clearly bounded task, not a system that takes over all customer contact.
Lead time hangs mostly on the connections. A chatbot on your own content ships faster than an agent allowed to write into three systems, simply because there is more to verify.
Frequently asked questions
What is the difference between an AI agent and a chatbot? A chatbot answers, an agent acts. A chatbot replies from its own content or from a connection that only reads. An agent decides which step is needed and performs it in another system, within the tools you have set up for it.
Is an agent always better than a chatbot? No. For questions with a fixed answer an agent only adds risk. The more the system is allowed to do, the more you have to arrange around permissions, review and escalation. Choose by task type, not by what sounds newest.
Can an AI agent write into our CRM? Only if you have set that action up as a tool. The model decides when to call a tool, not which tools exist; you write that contract. Start with read access and expand once the review step works.
What is a multi-agent system? Several agents that each handle part of the work and call on each other. That helps when the tasks genuinely differ, for instance retrieving and drafting. For most small business tasks one agent with a few well described tools is enough and easier to follow.
Looking at a demo and unsure whether it is a chatbot or an agent? Book a call. What runs at other companies is shown in the cases.