Module · 2026-10-30
Your first agent loop: one tool, one job
Before you build anything with code, you can run a real agent loop by hand this week, with one small job, one read-only tool and a clear way to check the result.
Starter module
What changes when a chatbot becomes an agent
A chatbot answers from what it already knows and what you paste in. An agent is given a job and at least one tool, and it decides for itself when to use that tool, looks at what comes back, and then decides what to do next. That repeating cycle of deciding, acting and checking the result is the agent loop.
The loop has five parts. You give the job. The model decides whether it needs the tool and says exactly what it wants from it. The tool runs, and the result goes back to the model. The model either asks for the tool again or says it has finished.
The important point for a business owner is that the model never touches your systems directly. It only asks for a tool to be used, and something else runs that tool and reports back. In a coded setup that something is a program. This week it will be you, which is the safest way to see how the loop behaves before you let a program do it.
Every loop also needs a stopping rule. Guidance on building agents from one major AI provider recommends a limit such as a maximum number of steps, so an agent that is stuck cannot keep going round in circles. You will set one in your brief.
Members only from here
The rest of this module (5 more sections, a hands-on exercise and a quiz) is for members. Membership is £19 a month or £149 a year, and you can cancel any time.