Andrej Karpathy, former OpenAI founding member and the person who coined “vibe coding,” built an OpenClaw agent that replaced six separate apps controlling his home. He described the experiment on a recent episode of the No Priors podcast, and Business Insider published an analysis on April 1 arguing that it illustrates how AI agents could dismantle the app economy.
What Karpathy Built
Karpathy’s agent, which he calls “Dobby” after the Harry Potter house elf, runs on OpenClaw and handles his home’s entire connected device stack through natural language. With minimal prompting, Dobby scanned his local network, identified devices, located undocumented APIs, and started handling commands — playing music, controlling lights, managing security systems.
“I used to use six different apps, and I don’t have to use these apps anymore,” Karpathy said on the podcast, according to Business Insider. “Dobby controls everything in natural language.”
The setup replaced individual apps from Sonos, lighting, and security systems. Instead of context-switching between vendor-specific interfaces, a single persistent agent handles discovery and control across all of them.
The App Economy Argument
Business Insider’s Alistair Barr framed the experiment as a signal that the organizational unit of software is shifting from “app” to “agent.” The thesis: if a personal agent can handle what previously required separate apps for music, home control, and task management, the downstream effects hit SaaS business models and app store economics.
This isn’t a prediction about apps disappearing tomorrow. It’s about what happens to app-level revenue when users no longer need to interact with individual vendor interfaces. If an agent mediates the interaction, the app becomes infrastructure — still running, but invisible and less monetizable.
Why Builders Should Pay Attention
Karpathy carries outsized signal weight with the builder community. His prior experiments — from vibe coding to his public AI workflow documentation — tend to show up in production setups within weeks. The Dobby experiment is specifically relevant to OpenClaw builders because it demonstrates a concrete use case: a single agent that replaces a fragmented home software stack by reverse-engineering local device APIs, rather than requiring vendor-provided integrations.
The technical skill required is still significant, as Business Insider notes. But the gap between “technically possible for Karpathy” and “one-click setup for everyone” has been closing fast in the OpenClaw ecosystem.