Most businesses buy the wrong AI tool. Not because they made a bad decision, but because the labels vendors use don't mean what you think they mean.

"AI chatbot", "AI agent", "AI assistant" — these three terms get used interchangeably in sales decks, product demos, and conference talks. They're not the same thing. And if you go into a procurement conversation without knowing the difference between an AI chatbot and an AI agent, you'll either overpay for something you don't need or end up with something that can't do the job.

This post breaks down what each one actually is, what it can and can't do, and how to figure out which one your business actually needs.

Why the terminology confusion costs businesses money

We've had this conversation more times than I can count. A business owner comes to us saying they want a chatbot. We ask a few questions about what it needs to do. Turns out they need an agent. Or the reverse: they've been quoted for a complex agentic system when a simple assistant would solve 90% of their problem for a fraction of the cost.

This isn't a small thing. The gap between a chatbot build and an agent build can be £10,000 to £50,000 depending on complexity. Getting the label wrong at the start means you're either massively overspending or setting yourself up for disappointment when the tool can't do what you actually need.

The problem is that vendors don't have much incentive to clarify. If they can sell you a more expensive "AI agent" when you only need a chatbot, they will. And if they can sell you a "smart AI assistant" that's really just a glorified FAQ widget, they'll do that too.

Knowing the real difference protects you in those conversations.

What an AI chatbot actually is (and what it can't do)

An AI chatbot is a tool that has a conversation with a user, takes their input, and returns a response. The core job is dialogue. That's it.

What it doesn't do on its own: take actions, query live systems, update records, make decisions, or chain tasks together. It responds. The user decides what to do with that response.

Scripted vs LLM-powered chatbots

There are two meaningfully different types of chatbot, and they get lumped together constantly.

Scripted chatbots follow decision trees. User asks X, bot responds with Y. Everything is pre-mapped. They're predictable, cheap to run, and break the moment someone goes off script. You've met these on every airline website and most bank support pages. They're frustrating for a reason.

LLM-powered chatbots use a language model underneath, something like GPT-4 or Claude. They can handle natural language, understand context across a conversation, and give much more nuanced responses. They're not following a script. But they're still fundamentally reactive. They respond to what you say rather than taking independent action.

The distinction matters because a lot of vendors are now selling LLM-powered chatbots as "AI agents". They're not. Being powered by a good language model doesn't make something an agent.

Appropriate use cases

Chatbots work well when:

• The primary job is answering questions (FAQs, product info, policy queries)

• Conversations are relatively self-contained

• You don't need the bot to update anything or take action in other systems

• Volume is the problem, so you need to handle 500 support queries a day without 500 people

They work poorly when the user needs something to actually happen as a result of the conversation. A booking made, a record updated, a follow-up triggered. That's where the other categories come in.

What an AI assistant actually is

An AI assistant is a tool that helps someone do their job better. It's user-facing and task-oriented. Think of it as a very capable tool that a person is still driving.

The assistant might draft an email, summarise a document, pull data from a spreadsheet, suggest responses to customer queries, or help someone think through a problem. The human is still in the loop making decisions. The assistant amplifies their output rather than replacing the process.

ChatGPT used as a writing tool is an assistant. Copilot in Microsoft 365 is an assistant. A tool that sits inside your CRM and helps your sales team draft personalised outreach is an assistant.

This is where a lot of "AI for business" tools actually sit, even when they're marketed as agents. They're tools that make your team faster. That's genuinely valuable. But it's different from automation.

Where assistants stop and agents begin

The line is about autonomy and action.

An assistant waits for a human to give it a task. It completes that task and hands the output back. A human then decides what to do with it.

An agent can be given a goal, not just a task, and work out the steps to get there on its own. It can take actions: query a database, send an email, update a CRM record, trigger a workflow in another system. It can make decisions mid-process based on what it finds. It doesn't need someone to hold its hand through each step.

That's the real difference. Assistants help people work. Agents do work.

What an AI agent actually is

An AI agent is a system that can take autonomous action to complete a goal. It's not just responding to a prompt. It's executing a process, often across multiple tools and systems, with minimal human intervention.

Here's a concrete example. A business receives a supplier invoice by email. An agent can read that email, extract the invoice data, check it against the purchase order in the ERP system, flag any discrepancies, update the accounting software, and notify the finance manager only if there's an issue. Start to finish, without anyone touching it.

A chatbot can't do that. An assistant can help someone do it faster. An agent does it.

How agents take action, not just respond

What makes an agent an agent is tool use. The system has access to external tools, APIs, databases, other software, and can decide which tools to use and when, based on what it's trying to accomplish.

At AMPL, we build agents using Claude Code rather than no-code platforms. The reason is that real business operations are messy. Off-the-shelf tools hit their limits fast when you've got a non-standard CRM, a legacy system that doesn't have a clean API, or a workflow that has 12 conditional branches. Custom builds handle that. Templates don't.

The key characteristics of a genuine AI agent:

• Given a goal, not just a prompt

• Can access and interact with external systems

• Makes decisions during execution, not just at the start

• Can chain multiple actions together in sequence

• Operates with minimal human intervention once triggered

Why agents require more build complexity

Agents are more powerful, but they're also more involved to build and maintain. There are a few reasons for this.

First, you need reliable integrations. An agent that can't consistently read from your CRM or write to your database is worse than useless. It'll create errors you don't catch. The integration work is often where the complexity lives.

Second, you need good error handling. What happens when the agent hits something unexpected? It needs logic for that. A chatbot that gives a slightly wrong answer is annoying. An agent that takes a wrong action can cause real problems.

Third, you need oversight and logging. You need to be able to see what the agent did and why, especially in early stages. This isn't optional. It's how you catch issues before they compound.

None of this means agents aren't worth it. For the right process, the ROI is significant. But it does mean you shouldn't let a vendor sell you agent complexity when you actually need chatbot simplicity, or the other way round.

Chatbot vs agent vs assistant: comparison table

Decision table: which do you actually need?

Here's a practical way to think about it.

You probably need a chatbot if:

• Your main problem is volume of inbound questions

• The answers are mostly information, not actions

• You want to deflect support tickets or handle FAQs

• You need something live fast with low complexity

You probably need an AI assistant if:

• You want your team to work faster, not replace the work entirely

• The output needs human review before anything happens

• You're dealing with creative, analytical, or communication tasks

• The process varies enough that full automation isn't sensible yet

You probably need an AI agent if:

• You have a defined, repeatable process that currently takes staff time

• That process touches multiple systems or has multiple steps

• You need things to happen automatically, not just get suggested

• The volume or frequency makes manual handling expensive

I'd add one honest caveat here. A lot of businesses think they need an agent when they need a better process first. We've had audits where the problem wasn't that the work wasn't automated. It was that the underlying process was broken. Automating a broken process just breaks things faster. Sort the process, then automate it.

If you're trying to work out where your problem fits, our free process audit is a good place to start. We look at what you're actually trying to fix before recommending anything.

Common mistakes when buying AI tools based on vendor labels

A few patterns we see repeatedly.

Buying an "agent" that's actually a chatbot with API access. Some vendors hook a chatbot up to one or two APIs, maybe it can look up order status or check a calendar, and call it an agent. That's not wrong exactly, but it's not what most people mean when they ask for an agent. If the tool can't chain actions, handle decision logic, or operate towards a goal autonomously, it's a chatbot with integrations.

Buying an assistant when you need automation. Microsoft Copilot and similar tools are genuinely useful for making individuals more productive. But if your problem is that a process takes 20 hours a week of staff time and you want that process to run without staff involvement, an assistant isn't the answer. It'll reduce those 20 hours to 15. An agent can get it to two.

Underestimating the data requirements. Agents need clean, accessible data to work with. If your systems are siloed, your data is inconsistent, or you don't have API access to key tools, an agent build will be slow and expensive. An honest vendor will tell you this upfront. Some won't.

Over-engineering based on a demo. Demos are designed to impress. The real question is how the tool handles your specific, messy, real-world process, not a curated example. Always ask to see it handling edge cases before signing anything.

To be honest, the single most useful thing you can do before any vendor conversation is audit your own processes first. Know what you're trying to automate, what systems are involved, what the current cost is in time and money, and what outcome you need. That clarity makes it much harder for vendors to sell you the wrong thing. We've written more about how to audit your processes before buying AI tools if that's useful.

FAQ

What is the difference between an AI chatbot and an AI agent?

Four things, basically. Autonomy: chatbots respond to inputs, agents act towards goals. Task scope: chatbots handle single exchanges, agents chain multi-step processes. Tool use: chatbots generate text, agents interact with external systems and databases. Decision-making: chatbots follow patterns, agents make conditional decisions mid-process based on what they find.

What is an AI assistant for business?

An AI assistant helps your team work faster, drafting emails, summarising documents, pulling data, suggesting next steps. A human is still driving the process and deciding what to do with the output. It's about making people more productive, not automating the process away. Tools like Microsoft Copilot or ChatGPT used for business tasks sit in this category.

Can an AI chatbot handle customer service?

Yes, within limits. A chatbot works well for answering common questions, handling FAQs, routing queries, or deflecting basic support tickets. An LLM-powered chatbot handles natural language much better than a scripted one. But if your customer service involves looking up live order data, updating records, or triggering actions across systems, you're looking at an agent, not a chatbot.

How do I know if I need an AI agent or just a better tool?

Ask yourself: do I need a process to run automatically with minimal human involvement, across multiple systems, at scale? If yes, that's an agent use case. If you just need your team to do the same process faster or with less effort, a good AI assistant or even better software might solve the problem for a fraction of the cost and complexity.

Why do vendors call everything an AI agent?

Partly because it's a compelling term right now, and partly because the definitions are genuinely blurry. Some tools sit on the boundary. A chatbot with deep integrations does share some characteristics with a simple agent. The practical test is whether the tool can operate autonomously towards a goal across multiple steps and systems. If it needs a human to prompt each action, it's not really an agent regardless of what the marketing says.

What do AI agents cost compared to chatbots?

Broadly, chatbot builds start cheaper, potentially a few thousand pounds for a well-configured LLM chatbot on your site. Agent builds are more involved: custom integrations, error handling, testing across real workflows, and ongoing monitoring mean the investment is higher. The ROI case is usually stronger too. If an agent recovers 20 hours of staff time a week, the numbers work out quickly. The audit we run before any build makes that comparison concrete.

If you're trying to work out which category your problem falls into, or if you've been quoted for something and want a second opinion, that's exactly what the AMPL audit is for. We look at your specific processes, map what's actually worth automating, and tell you what type of system would solve it. No commitment to a build required. Book a free conversation at amplconsulting.ai.