WhatisanAIagent,andhowcouldabusinessuseone?

An AI agent is software that takes a goal, works out the steps to reach it, and carries them out using tools you give it, such as your inbox, calendar, CRM or files. A chatbot answers a question and waits for the next one. An agent can read an enquiry, look up the customer, check availability, draft a reply and put it in front of a person for approval. The useful ones are narrow, have clear limits and keep a person involved at the points that matter.

Agent, chatbot or automation?

These words get used loosely, so here is a working distinction:

• Chatbot. You ask, it answers. It works from what you type and what it has been given. It does not take actions in your systems. A website help assistant or a general AI chat tool is a chatbot.
• Automation. A fixed sequence of steps that runs when something happens. “When a form is submitted, add a row to the spreadsheet and send a confirmation email.” It does the same thing every time and does not make judgements.
• Agent. Somewhere between the two. It is given a goal and a set of tools, and it decides which steps to take based on what it finds. It can handle variety that a fixed automation cannot, which is also why it needs clearer limits.

In practice, many useful systems mix all three: a fixed automation that hands one judgement step to an AI model, then waits for a person to approve.

Examples for a small business

• Enquiry handling. Reads each new enquiry, identifies what the person wants, checks your price list or availability, and drafts a reply for a person to send.
• Quote preparation. Takes notes from a site visit, matches them to your service list, builds a draft quote in your template and flags anything it could not price.
• Supplier invoices. Reads incoming invoices, matches them to purchase orders and lists the ones that do not match for someone to check.
• Research briefs. Given a company name before a sales call, gathers public information and writes a one-page brief.
• Internal questions. Answers staff questions about your own policies and procedures by searching your documents and pointing to the source.

None of these replace a person. They take the gathering, sorting and first drafting off someone’s desk, so the person spends their time on the decision.

Where a person stays in the loop

The most important design decision is where a person checks the work. Some good default rules:

• Anything that goes to a customer is approved by a person, at least until you have a long track record.
• Anything involving money, such as refunds, payments or price changes, needs human sign-off.
• The agent can read more than it can write. Read access to a price list is low risk; permission to change it is not.
• Every action is logged, so you can see what it did and why.
• There is a simple way to switch it off.

These boundaries, along with which data it can see and who owns the result, should be agreed before anything is connected. The responsible AI page explains how I handle that.

What agents are bad at

Agents can misread an ambiguous request, invent a detail that sounds right, or keep going when they should have stopped. They are weaker when the task depends on context that is not written down anywhere, such as a long-standing relationship with a customer. The narrower the job and the clearer the rules, the better they perform.

How to start small

1. Pick one task with a clear start and finish. “Draft replies to booking enquiries” is better than “handle customer service”.
2. Run it by hand first. Use a general AI tool on real examples, with a person doing each step, to see whether the AI part is good enough.
3. Write the rules down. What it may use, what it must never do, when it should stop and ask.
4. Connect the minimum. Give it only the tools and data that task needs.
5. Keep approval on. Let it draft and a person send, and measure time saved and errors caught.
6. Widen slowly. Only reduce review once the evidence supports it.

I have built AI products with this kind of structure myself; The Counsel, an open-source legal assistant for England and Wales, is one example of an AI tool working within set sources and limits. For business work, see agents and automation, or the wider AI consultancy page for how projects are scoped.

If you have a task in mind that might suit an agent, get in touch. After a first conversation, the scope, fee and timeline are agreed in a written proposal, and data boundaries and review points are settled before anything is connected.