What is an AI agent?
November 2026
The short answer
An agent is AI that does a task, not just answers a question. You give it a goal, it works out the steps, and it uses tools like your email, files or business systems to get it done, checking back with you where it should.
Chat versus agent
When you ask ChatGPT or Copilot a question and it writes you an answer, that is a chat assistant. You do the next step yourself: copy the text, open the system, send the email.
An agent takes those next steps itself. It reads the job sheet, looks up the customer in your CRM, drafts the quote in your template and puts it in front of someone to approve. The difference is that it can act, not only write.
They are already in the tools you pay for
You do not need to build anything to meet one. In the last few months: Microsoft’s Copilot Cowork carries out tasks across Microsoft 365. Business Central has a Payables Agent that matches supplier invoices. monday.com, Asana and Atlassian all now sell agents inside their project tools. Xero can now issue an invoice from inside Gemini once someone confirms.
What an agent needs to do a good job
- A clear task. “Draft a reply to every quote request using our pricing and terms” works. “Help with sales” does not.
- Your way of doing it. The rules, examples and templates a new starter would need. This is what a skill is for.
- Access to the right systems, and only those. That access comes through connectors, which is what MCP is for.
- A person who checks. Anything that goes to a customer, spends money or changes records should stop for a yes.
What to watch
Agents act with whatever access they are given, so check permissions before you switch one on. And most of them are now billed per task in credits, not a flat licence. Set a spending limit first, not after the first bill.
Where we come in
We build agents around a task you already do every week, connect them to your systems with the least access they need, and put the approval steps in the right places.