What an AI agent is
An AI agent is software that can carry out a task from start to finish, not just answer a question. You give it a goal, and it works through the steps: looking something up, deciding what to do, and doing it.
A simple chatbot answers questions. An agent can answer the question and then act on it. A customer service agent, for example, might answer a query, look up the order in your system, start a refund if your rules allow it, update the customer's record and send a follow-up email.
That difference matters. A chatbot saves you typing replies. An agent can save you whole chunks of process.
It's worth being clear about the limits too. An agent only knows what you give it and can only do what you've connected it to. It can make mistakes, so for anything that matters, it should check with a person or pass the job over.
Jobs businesses use AI agents for
Customer service. Handling enquiries by email, chat or phone. Sorting the simple ones and passing the rest to a person with the full history attached.
Sales follow-up. Researching who an enquiry came from, drafting a personal reply, following up if there's no answer, and flagging the serious leads to you.
Reporting. Pulling figures from your accounts, website analytics and other systems into a regular report, and pointing out anything unusual.
Documents. Reading invoices, forms or contracts, pulling out the key details, putting them in the right place and flagging anything that looks wrong for a person to check.
Bookings. Managing a calendar, handling changes and cancellations, and sending reminders, all linked to your existing booking system.
Staff questions. Answering common questions about policies or processes, and sending anything else to the right person.
Build or buy?
You have three broad options.
- Off the shelf. Ready-made tools are quick to set up and need no technical knowledge. They tend to be generic, can be hard to fit around how you actually work, and are often charged per user.
- Custom built. An agent made around your workflow, using your terminology and connected to the tools you use. It takes more work to set up and needs someone who knows how to build it.
- A mix of both. A custom agent built on top of a standard platform. You get some of the speed of off the shelf with more control over how it behaves.
If your process is fairly standard, try an off-the-shelf tool first. If your work has its own way of doing things, its own language, or relies on several systems talking to each other, a custom agent usually fits better.
What goes into building one
Every agent has four parts, whether you build it or buy it:
- A language model. This is the part that reads, reasons and writes.
- Tools and connections. These let the agent actually do things, such as send an email, look up a record or update your calendar.
- Memory and knowledge. Information about your business, your prices and rules, and what's already happened in a conversation.
- The rules that tie it together. The logic that decides what the agent does next, when it uses which tool, and when it stops and asks a person.
The fourth part is where most of the real work is. Getting an agent to behave sensibly in the odd cases, and to know when to hand over, takes testing with real examples from your business.
How to get started
- Pick one job that takes up a lot of time and follows a clear pattern.
- Write down how it's done now, step by step, including the exceptions.
- Decide which steps the agent should do alone and which need a person to approve.
- Build it, test it on real examples, and only then let it loose on customers.
Start narrow. An agent that does one job reliably is far more useful than one that does ten jobs badly.
At 8Caps, a standard agent build is a flat £495 founding price, paid only once you're happy. If it needs more than the standard package, we price it in writing before any work starts. AI usage is pay as you go, paid by you at cost, with a spending cap you agree.
Next steps
Read more about our AI agents, or book a free AI audit for a call and a written list of what's worth automating first. If you'd rather just talk it through, get in touch.
Not sure whether you need an agent or a chatbot? Our guide to AI chatbots vs AI agents explains the difference.