I haven't touched this in years, but one time I made a little project[1] to analyze the people I was following on Twitter and recommend who I might want to unfollow based on their attitudes. People who posted negative stuff very frequently were at the top of my list to ditch; I don't need extra input pushing me toward misery. The first few runs were very illuminated, but not surprising, like "wow, now that you mention it, Joe does saw awful stuff approximately hourly".

I would love to have an agent that could apply those sorts of analyses to my data sources. In my case, I wouldn't want to filter out bad news, but unnecessarily nasty spins on it. I'd find that super valuable.

[1] https://github.com/kstrauser/judgish