AI agent vs chatbot: answering a question, or finishing the job
A chatbot answers questions in a conversation, usually from a fixed set of documents, while an AI agent takes a goal, decides its own steps, calls tools to read and change live systems, and checks its work until the task is done.
Checked
A chatbot and an agent can look identical from the outside: a text box and a reply. The difference is behind the reply. A chatbot retrieves and responds. An agent reads the account, takes the permitted action, confirms it worked and records what it did.
In Motiontic's thesis on the four levels of AI in a product, the bolt-on chatbot is L1, the level users learn to ignore. The agentic product is L4, where the user states an outcome and supervises. Most of the distance between them is tools, permissions and evals.
What separates them
Goal, not question
A chatbot handles one message at a time. An agent works toward an outcome across many steps.
Tools
An agent calls APIs to look up orders, change records, send messages or run code. A chatbot usually has none.
Control of the path
OpenAI's guide defines an agent as a model that controls workflow execution and chooses tools within guardrails, and states that chatbots are not agents.
Verification
An agent checks whether its action worked and retries or escalates. A chatbot's turn ends when it replies.
When a chatbot is enough
If the job is answering questions from a stable help centre, and every follow-up action is something the user does themselves, a chatbot with good retrieval is cheaper and easier to run. Google's design guidance notes that predictable tasks are often cheaper without an agent.
What it takes to go from chatbot to agent
Scoped tool access
Connections to the systems the agent must read and change, with limits per action.
Approval rules
Which actions run on their own and which wait for a person.
Evals on real cases
A test set from past conversations and tickets with known correct outcomes, rerun on every change.
Traces
A record of every step, so a wrong action can be explained and fixed.
A chatbot compared with an AI agent
| Chatbot | AI agent | |
|---|---|---|
| Unit of work | One message and reply | A goal pursued over many steps |
| Source of answers | Fixed documents | Documents plus live systems |
| Actions | None, or a link to a form | Tool calls within set limits |
| Who decides the path | Fixed script or one model call | The model, within guardrails |
| Measure of success | Deflection and answer quality | Tasks completed correctly, with no reopen |
| Place on the thesis ladder | L1 bolt-on chatbot | L4 agentic product |
Common questions
What is the difference between an AI agent and a chatbot?
A chatbot answers questions in a conversation. An AI agent works toward a goal: it decides its own steps, calls tools to read and change live systems, and checks the result before it finishes.
Is a chatbot with RAG an AI agent?
No. Retrieval lets a chatbot answer from a company's documents, but it still only replies. It becomes an agent when it can choose and call tools to act, and decides the steps of the task itself.
Can an existing chatbot be turned into an AI agent?
Often, yes. The conversation layer can stay. The work is adding scoped tools, approval rules, evals built from past conversations, and traces of every action.
Are AI agents more expensive to run than chatbots?
Per conversation, usually yes, because an agent makes several model and tool calls. The comparison that matters is cost per resolved task, since an agent finishes work a chatbot hands back to a person.
Do AI agents still use a chat interface?
Some do. Others run in the background, triggered by an event such as a new ticket or order, and report back when the task is done or a person needs to approve a step.
Related
Agentic AI vs Generative AI
Content from a prompt, compared with a model that plans, calls tools and finishes the task.
AI Agents vs Agentic AI
One system that decides its own steps, compared with the wider category of AI that acts.
Customer service agents
Support agents that look up, act and escalate inside Zendesk, Intercom and the systems behind them.
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
Four levels of AI in a product
Why most products stop at a chatbot, and what levels three and four take.
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.