I run about 45 scheduled agents for marketing and engineering across multiple products. Most of them are not always-on, they are scheduled jobs. Some run mornings,some hourly. That covers most of what I wanted from always on and many of them have some type of event listener that can trigger a run as needed (like background research on a client signup).

The part that took the work was shared state. We used GPTree for the knowledge engine. Every routine reads and writes from the same knowledge and anything public or high risk gets an approval card to continue.