Cheap agents are better than clever agents

When I put an agent to work, I start with a simple question: how often do I really need its help? An agent running all the time is not necessarily more useful. It can also create noise and ask me to review things that have not changed.

So I cap each routine. For emails, it is three times a day on weekdays. For scanning meeting recordings, it is once in the evening. I prefer those defined moments to continuous monitoring. In this setup, the cost stays close to zero. That is not a promise for every use case; it is the frame I give mine.

The title is a little provocative. I am not looking for less capable agents. I want agents that do a specific task at the right time, without repeating it unnecessarily. A limited routine also makes me decide what I actually expect from it.

I keep approval before sending. The point is not to delegate my judgment, but to save my attention for decisions. If a routine stops helping, I review its role before asking it to work more. I prefer quiet support to an agent that keeps itself busy.