Folks, guess where I am? Small hint: there’s a giant Ferris wheel behind me. Joking aside, today’s reflection is about two categories of agents that are taking shape.
The agents that use your computer
The first category is the graphical, “computer use” agents. These are the agents that operate the computer in your place: they click, they drag, they scroll, they jump around while you sit there and watch them do the work. Perplexity just came out with a new agent, and they even built a browser specifically for it, called Comet, which is a fork and should ship soon.
The agents you never see working
The other category is made of agents you don’t actually watch doing things, because they connect through network services, through APIs, through the various protocols coming out like MCP, and they hook into other services to get things done for you.
Take an example: I want to book a holiday on Booking. The computer use agent, the graphical one, opens the browser for you, and you literally watch the Booking page come up and make the reservation. Meanwhile you sit there wondering, “wait, did it finish? Who knows.” The network-service agent, on the other hand, is a little chat that stays self-contained. You talk to the agent and say “book me a holiday on Booking.” It connects to Booking’s APIs, to the network services, and comes back to you with the reservation, or with some questions.
I see a nice civil war brewing here, derby-style, inside artificial intelligence. I won’t tell you which side I’m on. Let me know what you think.
