“AI agent” is 2026’s most abused phrase. Vendors use it to mean everything from a chatbot to a spreadsheet macro, which makes it very hard to know what you’re being sold. Here’s the plain-English version, with no stake in impressing you.

The one-sentence definition

An AI agent is software that can be given a job rather than a script — it reads the situation, decides the steps, uses your tools (email, calendar, CRM), and asks a human when it’s unsure.

Chatbot vs automation vs agent

  • A chatbot talks. It answers questions. It doesn’t do anything afterwards — the doing is still on you.
  • An automation follows a fixed path. “When a form is submitted, add a row and send an email.” Brilliant for predictable steps, brittle the moment reality goes off-script.
  • An agent handles the messy middle. The email that’s half an order and half a complaint. The PDF where the reference number is in the wrong box. It reads, judges, acts — and escalates what it can’t judge.

What one looks like in a real business

On a live build we run for a UK supply-chain operator, a document agent watches the inbox, reads each shipping document, pulls out the fields that matter, checks them against the rules, and pushes clean data into their transport system — saving a conservative 30 minutes per document. The part that makes it trustworthy isn’t the AI. It’s the exception queue: anything the agent isn’t confident about goes to a human, with its uncertainty flagged. Humans keep the judgement; the agent absorbs the volume.

When you should NOT want an agent

If the task never varies, a plain automation is cheaper, faster, and easier to trust — you don’t need judgement to copy a form into a spreadsheet the same way every time. If the task is rare, a checklist beats both. And if nobody in the business owns the process, an agent will just do chaos more efficiently. Anyone selling you “agents” before asking how the work actually moves is selling the word, not the system.

The question that cuts through every pitch

Ask: “What happens when it’s wrong?” If the answer isn’t a clear human-review path, walk away. That single question kills most of the AI-hype market — and it’s the first thing we design in every build. If you want to know whether your business has an agent-shaped problem or an automation-shaped one, that’s exactly what the audit answers, fixed-fee.