A few months ago I ran 10 AI agents in parallel. By Monday evening I was as tired as I used to be on a Friday. That was hard to accept.
Agents can do in seconds what used to take days: query dozens of databases, pull everything from Slack, the CRM, Google Drive and email, and turn it into a report or a spreadsheet. I only review it and think about what is important.
But doesn't this overwhelm us? Even removing the slow, routine tasks puts us in an extremely exhausting mode.
And in practice it's not seconds, it's minutes. You want to send another agent for another thing, and another, and another, and all of a sudden you're running dozens of agents. It doesn't matter whether you use a terminal with Claude Code (as I do) or an app like Cowork or Codex. It's all the same.
How I deal with it
I decided to run only one or two agents at the same time and to switch instead. While they work, I either switch to another task in the same project (for example, preparing prompts for it) or to a different physical context (loading the dishwasher, cleaning up at home). Either way, I stay with one project at a time, and it doesn't exhaust me.
Yes, I get less done than with 10 agents. But this is a marathon, not a sprint. We don't need to complete every task we could do in this world. Only the most important ones.
The next level is management: how to delegate your work to agents as you would to people. That will be a separate article. I think it's something leading companies will work on over the next few years.
How many agents do you run at once before it stops paying off?
First published on LinkedIn. Discussion on LinkedIn