Karpathy: The App Store Is an Outdated Concept — The Era of Bespoke AI Software Is Here

A Custom Health Dashboard Built in 1 Hour

AI researcher Andrej Karpathy shared a widely-discussed thread on X about the future of highly bespoke AI software. To track an 8-week cardio experiment aimed at lowering his resting heart rate from 50 to 45 BPM, he vibe-coded a fully custom tracking dashboard — and completed it in just one hour.

His AI agent (Claude) reverse-engineered the Woodway treadmill cloud API to pull raw data, process and filter it, and build a web UI frontend. He had to fix a few bugs — a metric/imperial unit mix-up and a calendar date matching error — but the overall result was impressively functional.

The Death of the App Store

From this experience, Karpathy drew two core conclusions:

• The app store model is outdated: Browsing, downloading, and using a discrete set of apps feels increasingly wrong when LLM agents can improvise a fully custom app in seconds, tailored exactly to your needs.
• The industry needs AI-native interfaces: Every product and service needs to reconfigure as a set of sensors and actuators with agent-native ergonomics. A treadmill should not maintain a human-readable frontend — it should expose an API or CLI that AI agents can easily use.

From 1 Hour to 1 Minute

Karpathy acknowledged that 1 hour is impressive but is excited by a more ambitious target: doing it in 1 minute. Reaching that requires the AI to have rich personal context, gather additional data automatically, reference skill libraries, and maintain a collection of the user's custom apps and automations.

He concluded: the future is not a set of discrete apps you choose from, but services of AI-native sensors and actuators orchestrated via LLM glue into highly custom, ephemeral apps. The traditional app store concept, he argued, is increasingly outdated on its own.