Uber Adopts 'Software Factory' Model as AI Agents Drive 70% of Code Changes
The ride-hailing giant is moving beyond AI assistants to a system where autonomous agents are primary contributors to the codebase.
Uber has fundamentally restructured its engineering workflow by integrating agentic AI across its entire software development lifecycle, transitioning toward what the company calls a "Software Factory" model. This shift marks a departure from using AI as a simple autocomplete tool to employing autonomous agents that can independently manage complex coding tasks.
According to data presented by Uber, the scale of this integration is already evident in the company's version control systems. More than 70% of pull requests at Uber are now attributed to local or cloud agents rather than human engineers. This means the vast majority of proposed changes to the company's codebase are now initiated and authored by AI agents, which then undergo review and integration into the production environment.
The Architecture of a Software Factory
Uber detailed this vision during a presentation at the AI Engineer World's Fair. The "Software Factory" approach relies on a framework of managed agents and modular building blocks designed to automate the software development lifecycle (SDLC). Unlike traditional "copilot" tools that require a human to drive every line of code, these agents operate with a higher degree of autonomy, handling repetitive tasks and complex refactoring at a scale that would be impossible for a human team to manage manually.
Redefining Engineering Productivity
This transition represents a massive shift in how large-scale engineering organizations operate. By moving AI agents from the role of assistants to primary contributors, Uber is effectively decoupling the volume of code production from the size of its human workforce. The consequence for the industry is a potential redefinition of the software engineer's role; the focus shifts from the manual act of writing code to the higher-level tasks of system architecture, agent orchestration, and rigorous verification.
The Path Forward
As Uber continues to scale this model, the industry will be watching to see how this affects long-term code maintainability and the onboarding of new human engineers into a codebase largely authored by machines. While the 70% pull request figure demonstrates immediate productivity gains, the long-term impact on software stability and the evolution of the engineering career path remains to be seen. For now, Uber's model serves as a blueprint for how the next generation of enterprise software may be built—not by humans aided by AI, but by AI managed by humans.