This feature is an experimental one at the moment, but it can produce good results. Basically, you should write a prompt into “Goal of the agent”. Columns are scraped fields. I know you are familiar with building custom scrapers. So this feature can build a scraper model for every page using AI vs you selecting fields on a page manually. You just give it a short context and set the fields to pull. It will output the same as any scraper.
For the use case you shared I would try setting “extract the url of the main image of musicians/band on the page” as goal and “Image url” as column.
Let us know if you have any other questions!
Victoria