Karpathy Demos Dobby: OpenClaw Agent That Reverse-Engineers Home APIs, Replacing Dozens of Smartphone Apps

On April 1, 2026, former Tesla AI director and OpenAI co-founder Andrej Karpathy demonstrated Dobby, an OpenClaw-based AI agent that fundamentally reimagines how humans interact with their connected devices. Instead of juggling dozens of individual smartphone apps for home automation, Karpathy built an agent that autonomously scans local networks, discovers connected devices, reverse-engineers their undocumented APIs, and controls them entirely through natural language.

In a demonstration on the No Priors podcast, Karpathy showed Dobby discovering and controlling Sonos speakers, smart lighting systems, and security devices, all without pre-configured integrations or manufacturer SDKs. The agent uses network scanning to find devices, then probes their HTTP, mDNS, and SSDP interfaces to understand control protocols, building its own integration layer on the fly.

The name Dobby references the household elf from Harry Potter, a fitting metaphor for an AI servant that handles domestic infrastructure invisibly. Business Insider covered the demo extensively, noting that it represents a shift from the current model of one app per device to a unified conversational interface.

This demonstration is significant because it shows agentic AI moving beyond digital-only tasks into physical-world device control. The agent does not rely on pre-built skills or MCP servers. It dynamically reverse-engineers whatever it finds on the network. This represents a new class of zero-config agent capability that could disrupt the entire smart home ecosystem.

The demo also highlights OpenClaw growing prominence as the agent runtime of choice for AI researchers. With 300,000+ GitHub stars and endorsements from figures like Karpathy, OpenClaw is establishing itself as the de facto platform for autonomous personal AI agents.