Agentic Anonymous

I run several agents at once, all day, and I have started to pay attention to what I actually do in the gap while they do their work. It turns out to be a specific kind of gap: too short to leave, too indeterminate to plan for. So I fill it with something abandonable — usually another agent, which produces another wait. This ends up with me being the bottleneck in a queue of activities I impose upon myself, each one having its own time horizon stretching ahead.

I'll compare the agents to a process I understand well — sourdough fermentation, where I can set an alarm and really focus on something else — to argue that what makes a wait leavable isn't transparency, but having a model good enough to enable one to “leave the room”. I can watch every tool call an agent makes and still not know when it will stop; I do not see yeast yet I know how long they will work for.

I'll close on Taylor and his scientific management theory, which prescribed workers' rest as carefully as it prescribed their work, provided management decided both. Nobody manages my idle seconds; I do it myself, and I'm worse at it than Taylor.