AI agent
An AI agent is software that uses a language model to decide what to do about a request and then carries it out using tools connected to real systems.
What it means in practice
A customer writes: “I ordered on Tuesday, order 4412, can you send it to my office instead?”
A chatbot replies with your shipping policy and a link to change the address. An agent reads the order number, checks the order, sees it hasn’t shipped, changes the delivery address, confirms the new address to the customer and notes the change on the order. Nobody on your team touched it.
The difference is tools. An agent can call specific actions: look up an order, check a calendar, send a payment link, book a courier, raise a refund. A chatbot can only write a reply.
What an agent is made of
- A model that reads the request and picks what to do.
- Tools, which are specific permitted actions in specific systems.
- Context, usually looked up from your own content. See RAG.
- Limits on what each tool may do, and when a person takes over.
Where the limits live
If your refund ceiling is written only in the model’s instructions, a confused or determined conversation can talk past it. If the software that runs the refund enforces the ceiling, nothing above the limit happens, whatever the conversation says. When you evaluate a tool, ask where each rule is enforced.
In AeroChat, the merchant sets limits on refund amounts, discounts, delivery dates and when to hand over to a person. The system applies them itself. You describe the workflow in your own words, and AeroChat works out which tools to call and lists any keys you need to add for outside systems.
What the analysts say
| Forecast | Date |
|---|---|
| By 2029, agentic AI will resolve 80% of common customer service issues without human intervention, with a 30% cut in operational costs | Gartner, 5 March 2025 |
| Over 40% of agentic AI projects will be cancelled by the end of 2027, because of rising costs, unclear business value or weak risk controls | Gartner, 25 June 2025 |
| Of the thousands of vendors selling “agentic AI”, Gartner estimates about 130 offer real agentic capability. It calls the rest “agent washing”: chatbots and assistants relabelled | Gartner, 25 June 2025 |
Source: Gartner press releases, 5 March 2025 and 25 June 2025. These are forecasts, not measurements.
A test for the word
Ask a vendor what the product can change in your systems without a person. If the answer is nothing, you are looking at a chatbot. That can still be useful for questions with a fixed answer. It won’t reduce the work that reaches your team, because those requests are the ones that need something done.
Related terms
Read next
Describe one workflow and watch it run
Start your 7-day free trial. No credit card needed. Pick the request your team handles most, such as an address change or a booking, describe it in your own words, and see what AeroChat sets up from it.
On Shopify? Install AeroChat from the Shopify App Store.