When AI Can Do Anything, Knowing What You Want Is the Real Power
Tech analyst Ben Thompson argues agents will turn apps into behind-the-scenes suppliers, much as the web did to publications.
Stratechery · Apps, Agents, and Aggregation | Stratechery by Ben ThompsonTechnology analyst Ben Thompson had just watched an AI assistant build him a small recipe organizer, complete with a custom interface meant for him alone.
The throwaway project was disposable, he said, because it was effectively infinite. It prompted his larger claim about where power will sit when software can do almost anything.
"The ability to do stuff is becoming abundant," Thompson said. "The problem to be solved is not discovery but rather inspiration."
Thompson called it "a replay of what happened with websites." When publications needed printing presses and delivery trucks, publications were scarce. Once the web made distribution free, he argues, discovery became the scarce problem, and the companies that solved it gathered customers and gained power over suppliers.
Thompson argues agents are doing the same thing to action. If an assistant can operate computers, use apps and websites, and generate a one-person interface, doing the task stops being the hard part.
The hard part, in his telling, is knowing what a person wants done in the first place, what he calls inspiration. Thompson says whoever solves inspiration will "gain power over every entity that has things that need to be done."
"Once the user is focused on solving a problem," Thompson says, "every app and service required to do so is abstracted away into an implementation detail." A supplier left competing for scraps while the agent holds the relationship with the person.
Thompson was writing after Meta Connect, where Meta showed new devices, primarily glasses. He says those devices "are not AR or VR devices or even AI glasses" but "Muse delivery mechanisms" for Muse, Meta's assistant. Power, he argues, goes to the agent that holds the user's goal, not to the app it uses.
Questions people ask
What is an AI agent, exactly?
An agent is AI connected to a computer so it can act, not just answer. It can click buttons, fill forms, read files, run software, and keep working in the background, while the language model handles the flexible reasoning about what to do next. Thompson describes it as AI with access to a computer, using ordinary software to get things done.
What does he mean by inspiration?
Thompson means the starting idea, or grasp of a person's goal. If execution is cheap and any agent can book, buy, organize, or build, the advantage goes to whoever best understands the desire, has the context, and is trusted to act on it. It is less about search results and more about knowing what job to do.
What is aggregation theory?
Aggregation theory is Thompson's argument that the internet rewards whoever gathers demand. When supply becomes abundant, customers flock to one convenient discovery point, and that middleman gains leverage over suppliers, as Google did with websites and app stores did with apps. He now says agents could play the same role for doing tasks.