Emulated: The Data for Fully Autonomous Software Engineers and Companies — Joseph Wang
Emulated is building training data for AI agents designed to work as software engineers in real production environments.
Their key insight is that current training environments are too simplistic—they only simulate clean code changes, not the chaotic real-world work of keeping systems running when databases crash, servers fail, and problems arise during ongoing incidents. Instead, Emulated creates simulations of entire companies and cloud infrastructure, where an AI agent must handle everything from resource allocation and cost management to scaling services when failures occur. The founders have backgrounds in network infrastructure and believe that by combining domain expertise with realistic simulations, AI agents can learn the messy, complete work that actual infrastructure engineers do.
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
