When an AI agent is worth building — and when it isn'tWhen an AI agent is worth building — and when it isn't

Almost every business we talk to now asks about AI, and almost none of them arrive with a problem. They arrive with a category. That is understandable — the category is loud — but it is the wrong end to start from, and it is how companies end up paying a monthly fee for a chatbot nobody uses.

The question that works

Instead of "what could AI do for us", ask: what do we do repeatedly, by hand, that follows rules we could explain to a new employee in ten minutes? That question has a short, concrete answer in most businesses, and the answer is usually where the money is.

• Answering the same six questions by email, twenty times a week.
• Reading an enquiry and deciding which of three people it should go to.
• Copying details from a form into a spreadsheet, then into an invoice.
• Chasing a customer who booked but did not pay.
• Writing the same three-paragraph reply with two details changed.

Each of those is small. None of them is interesting. All of them are exactly what this technology is good at, and collectively they are often several hours a week of someone's attention.

Where it works well

The pattern that reliably delivers is narrow scope, real data, clear handoff. An agent that answers questions about your specific services, from your actual documents, and hands anything it is unsure about to a human with the conversation attached. It does not need to be clever. It needs to be correct about a small number of things and honest about the rest.

Where it does not

Anything where being wrong is expensive and being slightly wrong is hard to notice. Quoting bespoke prices. Making promises about delivery dates. Giving advice with legal or medical weight. In those cases the right build is an assistant for your staff, not a replacement for them — it drafts, a person sends.

An agent that says "I'll get someone to answer that" is doing its job. One that invents an answer is a liability with a monthly fee.

The part nobody budgets for

Connecting things. The value of automation is almost never in the model — it is in the fact that your website, your inbox, your CRM and your calendar stop being four separate places where the same information is typed. That integration work is the real project. The AI part is often the smallest piece of it, and any proposal that treats it as the whole thing is describing a demo, not a system.