# AI agent vs chatbot: what a small business gets

Source: https://cplt.tech/blog/ai-agent-vs-chatbot-small-business
Author: Stéphane Lepain
Published: 2026-11-17T06:59:07.670749+00:00
Tags: ai-agents, small-business, automation

---

A chatbot answers questions in a conversation. An AI agent carries out the steps of a task in your tools: it reads an email, looks up the order, drafts the reply and updates the record, choosing the next step as it goes. For a small business the useful question is which of your work needs answers and which needs steps done.

## How the companies that build them define it

OpenAI's [practical guide to building agents](https://openai.com/business/guides-and-resources/a-practical-guide-to-building-ai-agents/) puts it in one line: "Agents are systems that independently accomplish tasks on your behalf." The same guide places simple chatbots, single-turn LLMs and sentiment classifiers outside that definition, because in those the model answers and something else controls the steps of the work.

Anthropic's engineering article [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) draws a second line that matters more to a small team. In a workflow, the model and its tools follow "predefined code paths" that someone wrote in advance. In an agent, the model decides: Anthropic's words are that "LLMs dynamically direct their own processes and tool usage." Both count as agentic systems in Anthropic's vocabulary, and the article also says the word "agent" can be defined in several ways.

Keep that in mind when a sales page says "agent", and ask which of the three it means: a chatbot, a fixed workflow or an agent that picks its own steps.

## What changes for you

With a chatbot, a person asks and reads the answer, then decides what to do with it. You can check the answer before acting on it. That fits questions like "what is our return policy for this product?", ideally from an assistant that answers from your own documents and shows the page it used.

With an agent, the AI acts in your email, your customer records or your invoicing. Microsoft's November 2024 WorkLab [explainer on agents](https://www.microsoft.com/en-us/worklab/ai-at-work-what-are-agents-how-do-they-help-businesses) lists skills such as the ability to "send emails, create support tickets, or update records". You gain finished work, and you take on the risk that it is done before anyone reads it, so the questions you ask before buying change too.

Between the two sits the fixed sequence of steps. For example, an invoice arrives, the AI pulls out the supplier, amount and due date, and a person approves the payment. The order is decided in advance and the model does the reading. Anthropic recommends "finding the simplest solution possible, and only increasing complexity when needed", and adds that this "might mean not building agentic systems at all." I would start most small-business automation there.

## Five questions to ask before you buy an agent

1. Which tools can it change, and which can it only read? Start with read access plus drafts, and add changes once you trust the results.
2. What does it do when it is unsure? OpenAI's guide says an agent "can halt execution and transfer control back to the user" when it fails. Ask the supplier to show you that happening on one of your own cases.
3. Who sees the result before it reaches a customer or a payment? Name the person. The [article on which tasks to automate first](https://cplt.tech/blog/which-tasks-to-automate-with-ai-first) explains why I shortlist work with a named reviewer.
4. Can you see each step it took on a finished task? You need that to fix the cause when a result is wrong, as described in [what to do when your AI gives a wrong answer](https://cplt.tech/blog/what-to-do-when-ai-gives-a-wrong-answer).
5. What does a finished, checked result cost? Count the retries and the time a person spends checking, as in the [cost per accepted result](https://cplt.tech/blog/ai-pilot-cost-per-accepted-result) method.

## Where an agent earns its place

OpenAI's guide says agents suit work "where traditional deterministic and rule-based approaches fall short", and its examples involve judgement and exceptions. That matches what I would look for: work your team repeats, with a clear source, where a fixed sequence keeps breaking on the odd case. Give the model more freedom only at those points and keep the rest as fixed steps.

If you have work that crosses your inbox and your other tools, look at how I approach [business process automation](https://cplt.tech/services/business-process-automation), then bring a few real examples to a [free 45-minute call](https://cplt.tech/contact). I will help you sort which parts suit an agent, which run well as fixed steps and which stay with a person.

## Frequently asked questions

### Is an AI agent the same as a chatbot?

They are different. A chatbot answers in a conversation and a person takes the next step. An agent uses tools to carry out the steps itself, such as updating a record or sending an email, and chooses the next step as it goes.

### What can an AI agent do for a small business?

It can carry out steps in the tools you already use. Microsoft's examples are sending emails, creating support tickets and updating records. I would start with work where a person checks the result before it reaches a customer or a payment.

### Do I need an agent or a chatbot?

If your team mostly needs answers from your documents, an assistant that answers and shows its source is enough. If the work is a chain of steps across your tools, an agent or a fixed sequence of steps can do it. Anthropic recommends the simplest solution that works.
