Apps, Agents, and Aggregation
Apps, Agents, and Aggregation
Thompson starts by treating smartphone apps as ways of accomplishing jobs, then argues that a computer-equipped personal agent can perform those jobs and generate task-specific UI rather than forcing people to learn many apps. In his own example, the first recipe artifact was an unusable PDF; the second was a personal index of saved Instagram videos. His further request to read captions and watch videos was not finished, and he reports Instagram rate limiting. From that one experience he advances a wider strategy claim: whoever controls the agent’s persistent context and inspires tasks could mediate users’ access to many suppliers, as Meta and Microsoft seek to do in different markets.
We cannot infer production quality, repeat use, security or maintenance from the five-minute interaction. His ‘disposable’ describes personal-interface replacement, whereas Dru’s concern is that a prototype can anchor the approach to a shared production system. A useful follow-up would revisit this same UI after recurring use, edits and repairs, then compare a team-maintained implementation of the same need.