25 May 2026
I’ve been having a lot of fun putting an AI agent to work on an unglamorous problem: keeping up with the news.
There is simply too much happening in AI to keep up. To stay afloat, I subscribe to a handful of newsletters. They are excellent, and it comes to roughly an hour of reading a day altogether. That is an hour I cannot surrender at my desk. But I do have pockets of screen-free time: walking, exercising, chores, not to mention that listening to the news is my preference in the first place!
So I handed the job to OpenClaw, the viral open-source personal AI agent. The brief: take my newsletters, strip out the adverts and filler, and read me the news. The result is a ten-minute daily audio digest, delivered on Telegram.
An agent with access to your email would be a security issue. So OpenClaw can only access its own dedicated mailbox, read-only. It runs on an isolated machine with no route into our networks. If it ever misbehaves, the worst it can leak is a newsletter.
The clever part is interactivity. While listening, I can hold the voice button in Telegram and ask, "Tell me more about that article." It replies with a longer clip, a link, or a summary — and can file articles to a reading list for later if I ask it to.
How well does it work? Quite well. OpenClaw is young and wobbles when updated, but the parsing has been as reliable as anything I’d write by hand. It was also able to resolve any parsing issues on its own, though sometimes it needed a little nudge to do so.
And because this wouldn't be a post about AI without a twist: a more mature alternative, Hermes Agent, has since come to my attention that promises significant self-improvement capabilities. I’m exploring it, stay tuned to hear my experience with it in a few weeks!
Author: Petr Klus
