What is an AI agent?
You have probably heard "AI agent" a lot this year. Here is what it means in plain English, how it differs from the chatbots of a few years ago, and what one can sensibly do for a small business.
In one sentence
An AI agent is an AI model, such as ChatGPT, Claude or Grok, that can take actions with tools you give it, like reading a message, checking a calendar and sending a reply, rather than only chatting.
In more detail
A chatbot follows a script: if the customer types "opening hours", it sends the opening hours. An AI agent reads the message, understands what the person wants even if they phrase it oddly, and decides which tool to use to answer.
The tools are what make it an agent. Without them, an AI can only write text for you to copy. With them, it can look up your price list, see free slots and send the reply in the right chat.
An agent still needs instructions. It follows the notes and rules you give it, such as which questions to answer, what to say about prices and when to hand a conversation to you.
An AI agent is usually not a separate robot you have to buy. In most cases it is an AI app you may already use, such as Claude or ChatGPT, connected to the services where your work happens. The same app that helps you draft an email can, once it has the right tools, read a customer's question and answer it in the chat where it was asked.
It is worth being clear about what an agent is not. It is not perfect, it does not know your business until you tell it, and it can get things wrong while sounding sure of itself. That is why the best set-ups give it a narrow job, clear notes and a quick way to pass anything unusual to a person.
Chatbot, set reply or AI agent: the difference
| Question | Scripted chatbot | Set reply on a trigger | AI agent |
|---|---|---|---|
| How it decides what to say | Follows a fixed menu or keyword list | Sends the same saved text whenever the trigger happens | Reads the message and works out what the person wants |
| Copes with odd phrasing or two questions at once | Rarely | No | Usually, if your notes cover the topic |
| Can take a next step, like booking | Only if every path was built in advance | Only the one action it was set up for | Yes, with the tools you give it |
| Effort to set up | Building every path by hand | A few minutes per rule | Writing clear notes and connecting tools |
| Main risk | Customers get stuck in menus | The same reply goes out when it does not fit | A wrong answer if your notes are vague |
Spaces offers the last two. Routines send a set reply, which Spaces calls a snippet, when something happens, such as a keyword arriving or no reply after a number of hours. Routines are not an AI: they never write anything new. Your AI agent is the part that reads, thinks and writes.
The four things every AI agent needs
- A model. This is the AI itself, such as Claude, ChatGPT or Grok. It does the reading and the writing.
- Tools. These are the actions it is allowed to take, such as reading a conversation, checking a calendar or sending a reply. Without tools, it can only suggest text.
- Instructions. Your notes on prices, hours, tone, what it may promise and when to hand over. This is where most of the quality comes from.
- Limits. Rules outside the AI that it cannot talk its way past, such as how many messages an account may send in a day. Good limits mean a mistake stays small.
If any one of these is missing, the agent either cannot do the job or does it badly. Owners often spend a long time choosing the model, when the instructions and the limits make a bigger difference from one day to the next.
An example conversation
Here is how an AI agent might handle a typical message to a hair salon on Instagram. Names, prices and times are examples.
| Who | Message |
|---|---|
| Customer | hiya do u do balayage and how much roughly, also is sat any good |
| AI agent (checks the price list and Saturday's diary) | Hi! Yes, we do balayage. It usually starts from £120 depending on length, and it takes about three hours. This Saturday we have 9:30am or 1pm free. Would either of those suit you? |
| Customer | 1pm pls. its for my wedding in 3 weeks tho is that ok |
| AI agent | Congratulations! 1pm on Saturday is booked for you. As it is for your wedding, I have asked Jess, our senior colourist, to check the plan with you before the day. She will message you here this afternoon. |
| Behind the scenes | The agent booked the slot, tagged the chat as a wedding booking and handed it to Jess, because the salon's notes say special-occasion colour always needs a person. |
A scripted chatbot would probably have struggled with the first message, because it asks two questions at once in casual spelling. The agent understood both, used two tools and knew from the notes when to bring in a person. That combination of understanding, action and judgement about its own limits is what people mean by an agent.
What an AI agent can sensibly do for a small business
- Answer questions about hours, prices, location, parking, stock and delivery, in the chat the customer started.
- Book, move or confirm appointments, and check the details back to the customer.
- Take simple orders and repeat them back so nothing is misread.
- Collect what you need before a quote, such as photos, sizes, dates or postcodes.
- Follow up with people who know you and expect to hear from you, and stop as soon as they reply or say stop.
- Tag conversations as leads, orders or problems, so you can see what needs you at a glance.
- Pass anything unusual to the right person, with a short summary so the customer does not have to repeat themselves.
What an AI agent should never do
- Make up prices, availability or policies it has not been given.
- Agree refunds, discounts or compensation unless you have written clear rules for them.
- Give medical, legal or financial advice.
- Argue with an upset customer.
- Message people who have not contacted you and do not expect to hear from you, or anyone who has said stop.
- Carry on replying in a chat after a person on your team has taken it over.
Writing these down as rules in your agent's notes is the simplest safety step you can take. A good agent follows them consistently. Spaces also enforces the last two on its side, whatever the agent decides, because they protect your accounts and the people you message.
How to get started, step by step
- Pick one channel and one job. For example: answer Instagram messages about prices and bookings. A narrow start is much easier to check.
- Write your notes. One page is enough to begin: what you sell, your prices, your hours, how to book, how you like to sound and the topics that always go to a person.
- Connect your AI agent to the account. With Spaces, that means connecting your account and adding Spaces to the AI app you use. The setup page for each AI app shows the exact steps.
- Watch the first week closely. Read every conversation the agent has. Where it gets something wrong, the fix is almost always a clearer line in the notes.
- Widen the job slowly. Once you trust it on prices and bookings, add delivery questions, or a second channel.
You will see the agent's replies in the same chat as always, on your phone or in the app, because the messages go from your own account. You can step in and reply yourself at any time.
Common mistakes
- Giving it no notes. An agent with no information about your business will guess, and guesses can sound convincing.
- Letting it handle complaints. Upset customers want a person. Make complaints an automatic handover.
- Starting with outreach. Using an agent to message people who did not ask to hear from you is the quickest way to get an account restricted.
- Never reading the conversations. A weekly look at ten chats catches most problems early.
- Treating it as finished. Prices, hours and offers change, so update the notes when they do.
Why it matters for your business
- It can answer customer messages in minutes, at any hour, in your words.
- It handles questions phrased in ways a script would not recognise.
- It can do the follow-up steps, like booking a slot, not just reply.
- It lets a small team keep up with several channels without someone watching every app all day.
- It should always hand complaints, refunds and anything sensitive to a person.
How Spaces handles it
Spaces gives an AI agent the tools to work on your own WhatsApp, Instagram, LinkedIn, Messenger, X and email accounts: reading conversations and sending replies. Spaces keeps every account inside safe daily limits, so the agent cannot send faster than the platforms allow.
Spaces works with the AI agent you already use, including Claude, ChatGPT, Grok, Cursor, Codex, OpenClaw and Hermes, through one MCP server. If you later change which AI you use, your connected accounts and their limits stay exactly as they were.
Some things are refused outright, however the agent is prompted: the same cold message sent to many people, and any message to someone who said stop in that chat. Spaces lowers the risk to your accounts, but it cannot promise a platform will never restrict one, so it is still worth keeping the agent to people who expect to hear from you.
In short
An AI agent is an AI that can act, not only chat. For a small business, its best use is answering the people who already message you, quickly and in your words, while passing anything that needs judgement to you.
Start small, give it clear notes, keep the hard cases for people and let firm limits catch mistakes. A good next step is to read how MCP connects an agent to your accounts, and what to put in your agent's notes.
FAQ
Is an AI agent the same as a chatbot?
No. A chatbot follows fixed rules and menus. An AI agent understands the message, even when it is phrased oddly, and chooses which tool to use to answer it.
Which AI agents can I use?
Any that can connect to tools, including Claude, ChatGPT, Grok and Cursor. Spaces also has setup pages for Codex, OpenClaw and Hermes, each checked against that app's own instructions.
Will an AI agent replace my staff?
It takes the repetitive messages off them, such as prices, hours and bookings. People still handle the conversations that need judgement, care or a decision, and they usually have more time for them.
Do I need to be technical to use one?
No. If you can add an app to your phone and write a page of notes about your business, you can set one up. Most AI apps connect to Spaces with a sign-in in the browser.
Can an AI agent make mistakes?
Yes. It can misread a message or fill a gap in your notes with a guess. Clear notes, a firm handover rule for anything unusual and a regular look at its conversations keep mistakes rare and small.
Should customers know they are talking to an AI?
Many owners tell the agent to say it is an AI if a customer asks, and to offer a person whenever one is wanted. Being open about it tends to build trust rather than lose it.
What does it cost?
There are usually two parts: the AI app you use, and the service that connects it to your accounts. Spaces plans are on the pricing page.