AI Agents vs Chatbots: What's Actually Different (and Which One You Need)
A chatbot answers; an AI agent acts. The real differences in planning, tool use and risk - plus how to tell a genuine agent from a chatbot with better marketing.
In short: a chatbot answers; an AI agent acts. A chatbot responds to a message with information, following either a script or a language model. An AI agent is given a goal, decides its own steps, uses tools and systems to carry them out, and reports back — booking the appointment rather than telling you the opening hours. Agents are more capable and more useful, but they're also harder to build, harder to control and riskier when they're wrong. This guide sets out the real differences and which one your business actually needs.
The one-sentence difference
A chatbot produces a response. An AI agent produces a result.
Everything else follows from that. If the software's job ends when it has said something, it's a chatbot. If its job ends when something has changed in the world — a booking made, a record updated, a refund issued — it's an agent.
Side by side
| Chatbot | AI agent | |
|---|---|---|
| What you give it | A message | A goal |
| What it returns | A reply | A completed task, plus a reply |
| How it decides | Scripted rules, or a single model response | Plans multiple steps, then adapts as it goes |
| Tools and systems | Usually none, or one lookup | Calls calendars, CRMs, payment systems, databases |
| Memory | Often just the current conversation | Carries context across steps and sessions |
| When it gets stuck | Repeats itself or falls back to a menu | Tries another route, or escalates to a human |
| Typical build effort | Days | Weeks, plus ongoing supervision |
| Main risk | Frustrating the customer | Taking a wrong action in a real system |
Three generations, not two
Part of the confusion is that "chatbot" covers two very different things.
Rule-based chatbots came first: decision trees with buttons. Predictable, cheap, and obviously robotic. They're still fine for narrow jobs like tracking an order.
LLM chatbots use a language model to understand free-form questions and answer in natural language, often over your own documents. Much better conversation, but still fundamentally a question-answering machine.
AI agents add three things on top: planning (breaking a goal into steps), tool use (actually calling your systems), and iteration (checking whether the step worked and trying again if not). That's the leap.
A useful test: ask the vendor what happens when the customer says "yes, book me in for Thursday morning." A chatbot tells them how to book. An agent checks the diary, offers a slot, takes the booking and sends a confirmation.
Why this distinction suddenly matters
The industry moved from one to the other very quickly, which is why the vocabulary is a mess.
Deloitte predicted that 25% of enterprises using generative AI would deploy AI agents in 2025, growing to 50% by 2027. In customer service specifically, Gartner forecasts that by 2029 agentic AI will autonomously resolve 80% of common customer service issues without human intervention, and cut operational costs by 30%.
Daniel O'Sullivan, a Senior Director Analyst at Gartner, framed the shift this way: unlike traditional generative AI tools that simply assist users with information, agentic AI "will proactively resolve service requests on behalf of customers."
The practical consequence for a buyer is that a lot of products are now marketed as "agents" that are, on inspection, chatbots with a nicer interface. The table above is a reasonable checklist for telling them apart.
Which one do you actually need?
A chatbot is enough when the questions are repetitive and the answer is information: opening hours, delivery times, specifications, policy questions, "do you cover my postcode?". If your enquiries are mostly people wanting to know something, a well-built chatbot over good content solves it cheaply and predictably.
An agent is worth it when the customer wants something done, and doing it means touching a system: booking, rescheduling, quoting, chasing, updating a record, routing a job. Anywhere a human currently reads a message and then types the same information into another screen is agent territory.
Neither is right when the interaction is genuinely sensitive. In a SurveyMonkey study of 2,017 US adults in December 2025, 79% said they strongly prefer interacting with a human over an AI agent, and 89% said companies should always offer the option to speak to a person. Preference for a human was strongest for financial or billing disputes (85%) and data security or privacy issues (78%). Those are conversations to route to a person quickly.
What changes when you move from chatbot to agent
The technology is the easy part. Four things get harder, and they're what separates a demo from something you'd let near real customers.
Permissions. An agent that can act needs to be told what it may and may not do. "Can book appointments, cannot issue refunds over £50, cannot delete records" — written down, enforced in the system, not just in the prompt.
Auditability. When an agent takes ten steps, you need a log of all ten. If you can't reconstruct what it did and why, you can't debug it and you can't answer a customer complaint about it.
Escalation. Agents should be confident about when to stop. A good one hands over cleanly with context attached; a bad one keeps trying and makes it worse.
Data protection. An agent touching customer records is processing personal data across more systems than a chatbot does. Under UK GDPR you'll need to be clear about what's processed, where it goes and how long it's kept — the ICO is the reference point, and it's worth settling this before the build rather than after.
What each one costs you — in effort, not just money
Price lists are a poor guide here, because the software is rarely the expensive part. The real cost is in the work around it.
A chatbot costs you content. Its quality is capped by the accuracy of what it can look up, so the bulk of the effort is writing good answers to your common questions and keeping them current. Build time is typically days. Ongoing effort is modest — mostly reviewing conversations and filling gaps.
An agent costs you design and governance. Someone has to decide what it may do, connect it to systems safely, define what happens when it fails, and review what it did. Build time is typically weeks, and the ongoing supervision doesn't stop. That's not a reason to avoid agents; it's a reason to budget honestly for them.
This is why "start with a chatbot" is common advice and often right — not because agents are worse, but because a chatbot forces you to fix your content, which an agent will need anyway.
The vocabulary you'll encounter
A few terms come up constantly and get used inconsistently. Rough definitions, so you can follow a sales conversation:
- Agentic AI — the umbrella term for AI that plans and acts rather than just responds. Usually a synonym for what an AI agent does.
- Tool use / function calling — the mechanism by which a model triggers real actions in your systems. This is the technical dividing line between chatbot and agent.
- RAG (retrieval-augmented generation) — looking up your documents before answering. It makes chatbots accurate; it is not, on its own, an agent.
- Orchestration — coordinating several steps or several agents. Relevant once you're past one simple task.
- Human in the loop — a person approves an action before it executes. Sensible for anything consequential.
- Copilot — AI that assists a member of your team rather than facing the customer. Often the lowest-risk place to start.
If a vendor uses these interchangeably, that's worth noticing.
The honest summary
Agents are more powerful, and for most operational jobs they're what you eventually want. But they're not automatically the right first step. Plenty of businesses get more value from a narrow, reliable chatbot over accurate content than from an ambitious agent that occasionally does something surprising in their booking system.
The sensible sequence is usually: fix the content, add a chatbot where the questions are informational, then introduce an agent on one well-bounded task with clear permissions — and expand from there once you trust it.
If you're weighing up which of these your business needs, talk it through with us. If the specific job is answering your phone, our guide to AI receptionists and virtual receptionists covers what a voice agent does in practice and what it costs in the UK.