Public questions that don't change
What time do you open on Sundays?
Hours, rates, directions. What anyone can read and always gets the same answer.
AI agent vs chatbot vs AI assistant, in three sentences. A chatbot follows a script somebody wrote in advance. An assistant answers whatever you ask it in the moment. An agent has a fixed brief and knows when it's its turn. One question tells them apart: who takes the initiative, the script, you, or the brief?
The example in this guide is a made-up neighborhood gym.
| What you look at | Chatbot | Assistant | Agent |
|---|---|---|---|
| Who takes the initiative | The script | You | The brief |
| What it knows | The answers somebody wrote for it | What it knows out of the box and, if you give them to it, your documents | The same as the assistant, plus its instructions |
| When it acts | When you click an option it expects | When you ask it for something | When something arrives that matches its brief |
| If you step off the expected path | “Sorry, I didn't get that” | It tries: it understands the question even when it's new | If it isn't its job, it stays out |
| Example at a gym | The menu on the website: hours, rates, cancel | “Summarize the weight room rules for me” | You paste a review and the reply comes out with the gym's sign-off |
A chatbot, strictly speaking, is a tree of questions and answers somebody wrote in advance. It works very well when the questions are few and always the same. At Cedar Street Fitness, the chat on the website handles hours and rates without bothering the front desk.
Its limit shows up fast. A member types “I want to freeze my membership for two months” and gets “Sorry, I didn't get that. Please pick an option.” It isn't being clumsy: that wasn't in the script.
An AI assistant has no script. It understands the request even when it's new, and it writes the answer on the spot. The initiative is yours: if you ask for nothing, it does nothing, and tomorrow you'll ask for something else.
What changes from one assistant to another is what it knows. A generic one knows a lot about the world and nothing about your gym. One that has read your documents answers with your house rules and your rates.
An AI agent is an assistant with a fixed job and written instructions: what it does, when it's its turn, how it does it, and what it can't promise. You no longer explain it every time. A review comes in and it knows that one's for it.
That's why the AI agent vs assistant difference isn't in the technology. It's in the brief: the same assistant, once you give it a fixed job, becomes an agent.
You'll also run into “agentic AI.” It's the trade term for software that plans and carries out several steps toward a goal, instead of answering one request at a time. In this guide's terms: the more the initiative sits with the brief and the less with you, the more “agentic” it is.
Two different things get sold under the word “agent,” and it helps to know which one you're being offered:
Almost always, and in this order: an assistant that knows the company and, on top of it, two or three agents for what repeats the most. The scripted chatbot still has its place on the website, for what's public and doesn't change.
Take Cedar Street, with a front desk, trainers and a chat on its website. It would end up with three layers: the scripted chatbot for hours; an assistant that knows the house rules, rates and protocols, for the team; and one agent for reviews and another for cancellations. How to set up the first one is in how to create an AI agent.
If all you want is to answer hours and prices on the website, a scripted chatbot is enough. The distinction matters when questions stop being predictable, or when a job repeats so often that explaining it each time wears you out. And if all you want is a saved request one click away, that's a shortcut: you don't need an agent.
The definitions in this guide are a practical rule of thumb, not a standard: every vendor uses the words its own way. All three are explained, one by one, in the glossary.
What we say about ChatGPT in the FAQ is our own reading, not a definition from OpenAI. How OpenAI presents it is in its help center.
Read each situation and decide. No trick: watch who takes the initiative.
What time do you open on Sundays?
Hours, rates, directions. What anyone can read and always gets the same answer.
I want to cancel my membership
A menu that routes to the right form or phone number, without interpreting anything.
What does the protocol say if someone gets dizzy in class?
It answers with the document in front of it and lets you see where the answer comes from.
Summarize the weight room rules in five lines
Today a summary, tomorrow a sign for the locker room. You ask for a different thing each time.
Two stars: “no bikes free at peak hours”
You paste the review and the reply comes out in the gym's tone and with its sign-off, with nothing to explain.
I'd like to cancel in March
It prepares the reply with the notice period, the steps and the option to pause the membership. Complete every time.
Flipa's agents are the second kind: they work inside your team and never talk to your customers. They prepare the text, and someone on the team sends or posts it. You don't have to remember which agents exist: paste the review and the one with that brief steps in. Under the answer it says which one it was.
Here's the reply, in the gym's tone. I checked the booking rules.
A chatbot follows a script: it only answers what somebody planned for. An AI agent understands new requests and has a fixed brief, with its own instructions. Step off the script and the chatbot is lost; the agent, if the request is in its lane, gets to work on it.
They're the same base, used differently. You ask an assistant for what you need each time. An agent is that assistant with a fixed job: what it does, when it steps in and what it can't promise.
By this guide's rule of thumb, an assistant: it doesn't follow a script and you take the initiative. It also has features for long, multi-step jobs, which come close to what we call an agent here. Their names change often: check OpenAI's site.
First an assistant that knows the company, because without that no agent gets it right. Then two or three agents for what repeats the most: reviews, replies to customers, social posts. Start with the agents and you'll get tidy briefs and generic answers.
It's the trade term for software that plans and carries out several steps toward a goal instead of answering one request at a time. The practical question stays the same: does it execute out there, or does it leave the work ready for a person to review and send?
The job you're most tired of repeating is usually a good candidate. If you want to try it in Flipa, start free, no card needed, and know that the app's interface is in Spanish today.
Start free