Karpathy proposes the concept of "Organization as Code": IDEs will become larger, and human programming will be upgraded to multi-agent team management.

Gate News reports that on March 12, OpenAI co-founder Andrej Karpathy introduced the concept of “organization as code” in a tweet today. He believes that in the era of intelligent agents, IDEs will not disappear but will become larger. Human programming units will shift from single files and individual agents to teams composed of multiple agents. Karpathy cited classic organizational charts from Amazon, Google, Meta, Microsoft, Apple, and Oracle, pointing out that these organizational models are essentially “organization code.” In the future, IDEs will help build, operate, and manage these models. He stated, “You can’t fork a traditional organization (like Microsoft), but you will be able to fork intelligent agent organizations.” He further argued that human organizations are opaque to CEOs, who cannot see the real-time status and data of every activity within the company; however, intelligent agent organizations are naturally observable and controllable in real time. He wrote, “I have no doubt that in the future, organizations can be managed via voice on a phone, with real-time statistical data.” Recently, Karpathy has been actively discussing multi-agent collaboration: on March 6, he released autoresearch (allowing AI agents to autonomously run GPU training experiments), on March 9, he launched AgentHub (an agent collaboration platform), and on March 11, he proposed the need for an “Agent Command Center” IDE. Today, he abstracted these practices into the “organization as code” conceptual framework, forming a complete evolution from single-agent experiments to multi-agent collaboration platforms, agent command centers, and the organization as code.