Ben Thompson argues AI agents need their own computers to become useful, and calls custom software the “infinite app”

Ben Thompson is framing AI agents as something more concrete than a smarter chatbot. In a new Stratechery post, he argues that “agents are not just AI” but AI with access to a computer, and that this distinction matters if the category is supposed to work for ordinary users.

Thompson’s point is that an agent becomes broadly practical only when the user does not have to assemble the underlying setup alone. He writes that to deliver an agent usable by people “unable or unwilling to set up a computer for their agent,” companies have to provide not only the AI itself but also “a computer for their agent.”

That idea sits at the center of his larger argument about how software may change as AI systems begin to operate computers on a user’s behalf instead of merely answering questions. Rather than treating apps as fixed destinations that people learn to navigate, Thompson is describing a model in which software can be created or reshaped on demand around a specific task.

As his example, Thompson describes asking Muse to organize recipes he had saved on Instagram. The result, he writes, was a recipe box app built for his own use. He then uses that small, personal tool to make a much bigger claim about where AI interfaces could be heading.

From personal tool to “infinite app”

Thompson writes that he initially thought of the Muse-built recipe box as just another app on his phone, but then reconsidered what it represented. In his telling, it was really an “infinite app”: an example of an agent creating any app he wants on command, even if that software is meant for just one person.