The smart home problem

At the end of my last post I said there’s a cognitive cost to managing workflows and agents. I’ve noodled on that a bit …

One of the risks with agents is the same problem I ran into with smart homes (aka IoT).

You start off getting one light bulb and it’s great: you hook it up, it turns on, it turns off. Then you have a few more things: a couple more light bulbs; maybe a robot vacuum cleaner; a couple of switches; some security devices; a couple of cams; and the doorbell.

Suddenly you’re managing 10, 15 things with schedules, with timers, with software updates, some things going out of support, some things you’ve forgotten the passwords for. What was a nice improvement (a light bulb turns on when you walk in the house) suddenly is a nightmare of things to manage.

It’s not made your life easier and the smart home is a real mission. You start to get really fed up, really fast.

I’m seeing the same thing with AI and agents. You start off and you find one process and you make it efficient. You’ve got one schedule, one agent running, and then you’ve got two and then you’ve got ten. Now you’ve got a problem managing the agents.

This comes at a cognitive cost.

Obviously the solution is more agents: agents managing agents, managing agents. Which starts to sound a bit like a workplace hierarchy with employees.

So in the same way employees need management and support, agents start to need management, time and energy.

I think one of the risks with agentic AI is the cognitive pressure of how many things you suddenly need to manage. It very quickly goes from agents saving you time to you spending that time managing the agents.

And similar to my conclusion in my last post, I think there’s going to be quite a long period before this settles down and we see how it lands. Does it become a solved problem, or a whole new problem space?

Related: A slow ceding of agency by Adam Singer, or what happens when you cede agency to agents. Another one to ponder and come back to.