Building a software factory for AI SDK (vercel.com)

The AI SDK team has developed a new software factory to manage the overwhelming influx of issues and pull requests it receives, largely due to the growing popularity of the project, which boasts over 20 million weekly npm downloads and more than 26,000 stars on GitHub. With a backlog exceeding 1,000 open issues and around 800 pull requests, the team recognized that traditional approaches to maintain the codebase were insufficient. Instead of solely relying on human contributors, they implemented a software factory that automates the lifecycle of pull requests and issue management, allowing agents to autonomously process incoming requests while still maintaining human oversight for quality assurance. The new software factory, referred to as ai-sdk-factory, automates specific tasks such as bug reproduction, implementing features, and reviewing changes, significantly improving efficiency. With a current capacity of generating 25-35% of merged pull requests and closing over 75% of issues monthly, it has helped reduce open issues from 1,022 to 844 in just a few weeks. The approach combines human involvement with extensive automation tailored to manage risks effectively, ensuring that while agents expedite processes, the final say remains with human reviewers, thereby maintaining quality and security in a foundational AI project.