Uber Builds AI Software Factory With 3,600 Agent Skills

Uber Engineering is moving AI agents beyond individual coding assistance and into the execution layer of its enterprise development workflow. Its approach uses a multi-agent architecture to divide work and packages standard operating procedures into reusable Agent Skills. The system is designed to let agents follow consistent processes, coordinate tasks and increasingly take ownership of software work, advancing Uber’s vision of an automated software factory.

Uber said more than 70% of its pull requests are now attributed to agents operating locally or in the cloud, underscoring the scale of adoption across its engineering organization. The company has also built about 3,600 Agent Skills covering different development scenarios and workflows. The figures show that Uber’s focus has shifted from isolated code generation toward coordinated agents that can execute SOP-based tasks and deliver work through established enterprise processes.