Vendors now describe almost every conversational product as an "AI agent," including tools that are unchanged decision-tree chatbots underneath. The label has become close to meaningless in marketing copy, which is a problem when you are the one signing the contract. This article defines the difference precisely, in plain language, and gives you a short test you can run on any vendor demo.
It is written for founders, marketing leads, and operations managers evaluating options, not for engineers. If you want the technical layer underneath, our guide to RAG knowledge base chatbots covers the retrieval architecture in depth.
The short answer
A chatbot follows a script that a human wrote in advance. Someone anticipated a question, wrote the answer, and connected the two. When a visitor asks something outside that script, the bot falls back, usually to a contact form.
An AI agent works from a goal and a set of tools rather than a script. It can look up information in your systems, reason about what the person actually needs, take a permitted action, and decide when a human should take over. Nobody wrote the specific answer in advance, because the answer depends on data that changes.
Why the distinction matters commercially
The practical consequence is maintenance cost and coverage. A scripted bot degrades quietly: your catalog changes, your policies change, your pricing changes, and the script does not, so the bot becomes progressively more wrong until someone notices. Keeping it accurate is an ongoing content job that nobody enjoys owning.
An agent connected to live systems stays current automatically, because it reads the source of truth at the moment it answers. That is the real return: not that the AI is smarter, but that it does not drift away from reality between updates.
The trade-off is governance. A scripted bot cannot say anything you did not write, which is limiting but safe. An agent can compose new sentences, which means you need explicit policy rules, an audit trail showing which record produced each answer, and clean escalation to humans. We break down those three controls in our guide to deploying a Shopify AI agent, and the same principles apply to any industry.
Four questions that expose which one you are being sold
- "What happens when someone asks something you did not anticipate?" If the honest answer is a fallback form, it is a chatbot.
- "Which of my systems does it read from, in real time?" Vague answers here mean the knowledge is static.
- "Can it take an action, or only reply?" Looking up an order, capturing a lead, tagging a record, and routing to a person are agent behaviours.
- "Show me why it gave that specific answer." If there is no trace back to a source record, you cannot debug or defend the output.
When a chatbot is genuinely the right choice
Not every situation needs an agent, and it is worth saying so plainly. If your questions are few, stable, and identical every time, a scripted bot is cheaper, faster, entirely predictable, and easier to approve with a legal team. Clinic opening hours, a booking flow with fixed steps, or a short qualification questionnaire are all well served by a decision tree. Complexity you do not need is a cost, not a feature.
Agents earn their place when the answer depends on data that changes: inventory, orders, availability, account state, pricing, or anything else that would make a written script wrong within a week.
What this looks like in a real product
To make it concrete: ConvertPilot is an AI agent rather than a chatbot because it reads live product, order, and policy data at the moment a customer asks, recommends items from current inventory, retrieves real order status, captures leads into your CRM, and escalates to your team with the full conversation attached. Every reply carries an audit trail showing the record behind it, and merchants set hard rules for discounts, shipping, returns, and tone. For ecommerce specifically, ConvertPilot for Shopify installs directly from the Shopify App Store with a free plan.
For the multi-channel version of the same architecture, see how ConvertPilot unifies chat, email, WhatsApp, and social into one AI inbox. If you are evaluating the broader market, our AI customer support automation guide and AI agent development company guide cover selection criteria and build-versus-buy in more depth.
Choosing well
Start from the questions your customers actually ask, not from the technology. List the top twenty, mark which ones depend on data that changes, and count. If most of them are static, buy the simple thing. If most of them are live, you need an agent, and your selection criteria should be grounding, guardrails, and auditability rather than model names.
If you would like a second opinion on that list before you commit budget, get in touch. We build both kinds of systems in our software practice, and we will tell you when the cheaper option is the correct one.