The words "chatbot" and "agent" get thrown around as if they mean the same thing. They do not. Getting this wrong is expensive, because you can spend months building the wrong tool for the job.
We build both for UK businesses, so here is the honest version, without the marketing gloss.
The short version
A chatbot answers questions. An agent does work.
A chatbot listens to what a customer types, finds the best matching answer, and replies. It is a conversation layer over your information. Ask it your opening hours or your returns policy and it will tell you.
An agent goes further. It can take actions on your behalf: book the appointment, raise the refund, update the CRM record, chase the supplier, send the follow-up email. It uses tools, not just text.
That difference sounds small. In practice it changes everything about cost, risk, and what you get back.
What a chatbot actually is
Most "AI chatbots" you see on business websites are one of two things.
The first is a scripted bot. You define the questions and the answers in advance, usually as a decision tree. It is predictable and cheap. It also breaks the moment a customer phrases something you did not anticipate.
The second is a retrieval bot powered by a language model. You feed it your website, your policies, your product pages, and your FAQs. When someone asks a question, it finds the relevant material and writes a natural answer. This is the modern version, and it is genuinely good at deflecting repetitive enquiries.
What a chatbot cannot do on its own is change anything in the real world. It talks. It does not act.
For a lot of businesses, that is completely fine. If most of your inbound messages are "do you deliver to my area" and "can I bring my dog", a well-fed chatbot handles them all day and night and frees your team up.
What an AI agent actually is
An agent is a language model with hands.
You give it access to tools: your booking system, your database, your email, your payment provider, your ticketing software. Then you give it a goal. It works out the steps, calls the right tools, and completes the task.
A concrete example. A customer messages at 9pm to reschedule a fitting. A chatbot says "please call us during opening hours". An agent checks the diary, offers three real slots, moves the booking when the customer picks one, and sends a confirmation. Nobody on your team touches it.
Another example. A supplier emails about a delayed order. An agent reads it, matches it to the affected customer orders, drafts the apology with a revised date, and flags anything worth a human eye.
That is the leap. From answering to doing.
Why the difference matters for your budget
A chatbot is cheaper to build and cheaper to run. It reads and writes. The risks are low, because the worst outcome is a wrong answer, which you can fix by improving the content it draws from.
An agent costs more, because every action it can take needs to be built, tested, and secured. If an agent can issue a refund, you need guardrails so it does not issue the wrong one. If it can update customer records, you need to be sure it does not corrupt your data. That is real engineering, not a plugin.
So the honest guidance is this. Do not pay for an agent when a chatbot solves your problem. And do not expect a chatbot to do an agent's job, no matter how clever the demo looks.
How to tell which one you need
Start with your actual bottleneck, not the technology.
If your team spends its day answering the same questions, you need a chatbot. The value is deflection: fewer emails, fewer calls, faster responses.
If your team spends its day on repetitive tasks that follow clear rules, you may need an agent. The value is completed work: bookings made, tickets resolved, data updated.
A quick test. Look at your last fifty customer interactions. If they mostly end with information being given, a chatbot covers it. If they mostly end with something being done in a system, an agent is where the return is.
Most businesses we work with start with a chatbot, prove it saves time, then add agent capabilities to the specific tasks that justify the extra build. That order keeps risk low and lets you learn what customers actually ask for before you automate the doing.
The bit nobody mentions: your content and your systems
Both tools are only as good as what sits behind them.
A chatbot needs accurate, up to date information. If your website says one thing and your policy says another, the bot will confidently repeat the contradiction. Cleaning up your content is often the real work.
An agent needs clean systems and clean data. If your booking system is a shared spreadsheet and three people's memories, no agent can navigate it reliably. Automation exposes the mess you have been getting away with manually.
This is why we treat these projects as engineering, not as a bolt-on widget. The AI is the easy part. Connecting it safely to the things that run your business is where the skill lives. If you want the deeper view on how we approach that, our engineering work covers it.
A sensible path forward
Here is what we would suggest for most UK businesses.
First, write down the top ten things people contact you about. That list tells you whether your problem is questions or tasks.
Second, tidy the information those questions rely on. This helps your website, your search rankings, and any bot you build later. It is useful work even if you never buy any AI.
Third, start with a chatbot scoped to those top questions. Measure how many enquiries it handles without a human.
Fourth, once you know your real patterns, pick one or two repetitive tasks and build an agent for those specifically. Narrow and reliable beats broad and flaky every time.
If you are weighing this up alongside your wider setup, our guide on where to actually start with AI automation pairs well with this one.
The goal is not to have the fanciest AI. It is to give your team back hours and give your customers faster answers. A chatbot does the first part of that today. An agent does the harder second part when you are ready for it.
Get the diagnosis right and the tool almost picks itself.
