Why the App Store Model Is Obsolete: Karpathy’s Radical Call for On‑Demand App Creation

For years the App Store slogan "There's an app for that" defined the mobile era, but the rapid advancement of large language models (LLMs) and autonomous agents is reshaping that landscape. In a recent interview, AI pioneer Andrej Karpathy claimed that future applications should be created on the fly rather than downloaded from a crowded store.

Karpathy illustrated his point with a personal experiment: while doing cardio, he used the Claude AI assistant to reverse‑engineer the cloud API of his Woodway treadmill and, within an hour, built a highly customized dashboard to track an eight‑week training plan. The process involved pulling raw data, filtering and debugging unit mismatches, and assembling a lightweight web front‑end, highlighting both the speed of generation and the practical bugs that still arise.

From this experience he concluded that the App Store model is "awkward and outdated" because a few hundred lines of code can now be generated by an LLM agent in seconds. He argued that the industry must shift toward sensor and actuator services that expose AI‑native command‑line interfaces (CLI) or APIs, allowing agents to invoke them directly without a human‑focused UI.

Karpathy lamented that 99 % of current products still rely on static HTML/CSS documentation and lack AI‑native CLIs, making them ill‑suited for instant, agent‑driven automation. He posed the challenge of reducing the end‑to‑end creation time from an hour to under a minute, envisioning a future where a simple spoken request like "track my cardio for the next eight weeks" would automatically assemble a temporary, purpose‑specific app.

The community response was mixed. Supporters praised the vision of LLM‑driven, highly customized applications, while skeptics warned that app stores provide a valuable safety layer and that most users lack the motivation or expertise to craft bespoke software. Some suggested an "App Store 2.0" that offers a base app with AI‑driven customization hooks.

Karpathy countered that the reliance on discrete apps reflects a scarcity mindset; as software becomes cheap and abundant, the notion of a permanent, downloadable app will fade, giving way to transient code paths assembled and discarded after a single use. He acknowledged that the transition will be gradual and mixed, but maintained that the underlying trend is plausible.