Comment by theshrike79

Agents run tools in a loop.

Yes, you can give the Big Brain Thing a vague task and expect results, some times it'll do it right

But if you want repeatability, give it tools to determine what is a "disruption".

I have this exact system running on a WhateverClaw with a few simple tools: weather, train time tables (I commute via train) and a read-only version of my "will I be at the office or remote today" -calendar via a tool. Oh and a tool to notify me via Telegram.

It gets this information using the given tools and determines if it's worth notifying me.

TBH this doesn't need a "claw", I could just run the tools in cron, construct a prompt with that data and run any LLM on it.