Chatbot and AI agent are not the same thing. Plenty of vendors would prefer you didn’t know that.
The confusion is understandable, because both get sold under the same banner of AI-powered customer service, and both arrive in your inbox wrapped in near identical case studies. But the products underneath do fundamentally different jobs, and with adoption accelerating the cost of muddling them up is growing. The ONS found that 29 per cent of UK businesses were using at least one form of AI by June 2026, rising to 49 per cent among firms with 250 or more employees. Customer service is one of the most common places that money lands. So it is worth being precise about what it buys.
Most businesses have already met a chatbot, usually in the corner of their own website. It is a scripted system that matches keywords in a customer’s message against a decision tree someone built by hand, then serves whatever response sits on that branch. Ask where your parcel is and it points you at a tracking page. Ask two things at once, or anything the tree never anticipated, and it stalls, loops, or quietly dumps you into a queue with an apology. For a narrow band of repetitive questions this still works fine, and it is cheap to run once built. Chatbots were designed for deflection, keeping enquiries away from human staff, and they were priced per seat or per message volume accordingly, with no reference to whether anyone’s problem actually got solved.
What a chatbot cannot do is act. It can recite the returns policy all day. It cannot process the return.
That is the line the current generation of AI agents crosses. Rather than matching keywords to a script, an agent works out what the customer is actually asking, poses a clarifying question if the request is ambiguous, and then carries out the task across the systems the business already runs on, whether that is order management, billing or the CRM. Amending a delivery address. Cancelling a subscription. Applying a refund and confirming it, all inside one conversation, on chat, email or increasingly voice. The industry calls this agentic AI, and the plainer way to put it is that the old technology answers while the new one resolves.
Follow the pricing and the difference gets even clearer. Because chatbots deflect rather than resolve, their cost has historically had nothing to do with outcomes, and a business could pay for a year of licences, watch the containment numbers look respectable, and still find most customers reaching a human anyway after a frustrating detour. Agentic platforms are increasingly priced per successful resolution instead. That is not a cosmetic difference. It ties the invoice to the result, which makes the technology far easier to weigh against the salaries and overheads it is supposed to offset. The analysts expect this model to win. Gartner predicts that by 2029 agentic AI will autonomously resolve 80 per cent of common customer service issues, cutting operational costs by around 30 per cent for organisations that deploy it well.
None of which means every proposal marked agentic deserves the label.
A few questions separate the two technologies quickly in a sales conversation. Does the system take actions in other software, or does it only produce answers? Ask to watch it complete a task, not describe one. How does it connect to your existing stack, and can it work with the systems you already have rather than demanding a migration? What happens when it fails? A genuine agent escalates to a human with the full conversation attached, so the customer never repeats themselves. And what oversight do you get, in terms of quality assurance, policy controls and a record of how the system reached its decisions? A vendor who cannot show you any of that is selling a chatbot, whatever the invoice says.
The stakes here are not abstract. The UKCSI, the Institute of Customer Service’s national barometer, reached 78.3 out of 100 in July 2026, its strongest reading in four years, driven largely by more customers saying their issue was handled right first time. That recovery took UK service teams two years of work and it is fragile. A badly scripted bot that traps people in loops chips away at exactly the trust being rebuilt, and British consumers hold grudges about that sort of thing. An agent that fixes the issue outright, or hands over gracefully when it cannot, protects it.
So when the next pitch lands promising AI-powered customer service, skip the question of whether it uses artificial intelligence. Nearly everything does now. Ask whether it responds or whether it resolves. That single distinction decides what your money buys, how the pricing works, and whether your customers end the conversation with their problem fixed or with a link to a help article they could have found themselves.
