An AI agent in one sentence
An AI agent is software that uses artificial intelligence to complete a task in several steps: reading an email, deciding what to do, updating a file, then moving on to the next one. It does more than answer a question.
Agent, chatbot, or classic automation?
- Classic automation: fixed rules (if X, then Y). Reliable and inexpensive when cases are predictable.
- Chatbot: it answers questions, for example on your website. It does not act inside your systems.
- AI agent: it chooses the steps and uses tools (email, files, software) to reach a goal, within limits and under supervision.
Examples for a small office
- Read a freight broker’s incoming emails and create a shipment record for each request.
- Read PDF invoices and enter the fields in your software, flagging anything that looks abnormal.
- Draft a reply to a quote request, which you review before it is sent.
- File incoming documents by client folder, with a list of exceptions to check.
When it is not a good idea
- The task is rare or different every time.
- A simple rule would do: classic automation costs less and behaves predictably.
- A mistake would have serious consequences and nobody could fix it in time.
- The data is sensitive and you have no plan to protect it (in Quebec, Law 25 governs personal information).
How to start
- Pick one specific repetitive task, with a beginning and an end.
- Write down how long it takes today.
- Try it on 20 to 50 real examples, including the difficult ones.
- Keep human review for sensitive actions.
- Count the mistakes before extending to other tasks.
Five questions to ask before hiring someone
- What happens when the agent gets it wrong?
- Where does my data go, and who can see it?
- Can I see a log of what the agent did?
- Who approves sensitive actions?
- How do we measure that it works, before and after?