What is an AI agent?
Software that holds a conversation, looks things up in your documents, and takes actions on your systems — deciding for itself which to do next and following a task through, step by step, rather than only replying to one message at a time.
A chatbot answers. An agent decides.
Both take a question and produce a reply, so from the outside they can look the same. The difference is what happens in between. A chatbot maps your message to a response — from a script, or from whatever a language model produces. An agent is given a goal, a set of tools, and a boundary, and it works out the steps itself: look this up, ask a clarifying question, call that system, then answer.
The three things an agent has that a chatbot doesn't
Tools it can actually call. Not "here is a link to our booking page" but placing the booking. An agent is connected to real systems and can act in them, which is why the interesting question about any agent is what it is permitted to touch.
A loop. It can take a step, look at the result, and decide the next step. If the first search returns nothing useful it can search differently. A chatbot has one shot.
A memory of the relationship. Not just the last few messages, but what it learned about this person the last time they spoke — and the time before.
"Agentic AI" is the same thing
The industry settled on agentic AI for this category — software that plans, uses tools and acts rather than only replying. If you arrived here searching for that, you are in the right place: the pages in this section describe an agentic system, built for one business rather than as a general assistant.
We tend not to use the phrase in our own writing, because it has been attached to enough products that it no longer narrows anything down. What narrows it down is the list above: tools it can actually call, a loop, and a memory of the relationship.
Where the word gets abused
"Agent" is now attached to almost anything with a text box, so it is worth being blunt about what does not make something an agent: being powered by a large model, having a personality, or being able to answer in seven languages. Those are all features of chat interfaces. If it cannot do anything except talk, it is a chatbot with better prose.
What an agent is made of here
Those three requirements are not abstract on this platform — each is a thing you configure, and the rest of this section is one page per piece:
| The agent… | Comes from |
|---|---|
| knows your business | two knowledge bases — your vetted material and each customer's own documents |
| behaves like your business | its soul, under a moderation layer no tenant can disable |
| follows your procedures | Skills, loaded only when the conversation needs them |
| reaches your systems | tools over MCP, enabled only through a loaded Skill |
| knows which page you came from | a page playbook |
| collects facts cleanly | ask forms it composes on the spot |
| acts in the gaps | scheduled tasks that fire months later, in your customer's time zone |
| keeps customers apart | a space each, enforced in the database |
If you are weighing this against a chatbot, the row that decides it is usually the last two.
The practical test: ask what happens after the conversation ends. If the answer is "nothing", you are looking at a chatbot.
The model is the engine that turns text into text; the agent is everything built around it — retrieval, tools, memory, boundaries and channels. Models are replaceable parts inside an agent.
An agent that works for one specific business — answering only from that company's own documents, staying inside a scope its owner defines, remembering each customer across visits, and accountable for what it says. Not a general chatbot, but a member of that business's team.
The engineering layer wrapped around a language model — retrieval, tools, skills, memory and guardrails — that turns a general model into a specific, dependable agent. The model supplies fluency; the harness supplies grounding, boundaries and the ability to act, and stays the same when the model is swapped.
Headless software exposes everything over an API, MCP and skills so that an AI agent — not only a human clicking a UI — can operate it. You describe the outcome; your agent, handed exactly the right skills, drives the features precisely.
A machine-to-machine channel where a customer's personal AI assistant talks directly to a business's agent — consulting it and submitting a structured request — instead of a person typing into a chat box. agent4.io handles the pre-processing half and hands back a summary, while a human confirms anything irreversible.