General Motors Triples Merged Pull Requests with AI Agents

General Motors' (GM) autonomous vehicle division has significantly boosted its engineering efficiency by integrating AI agents into its workflows. This strategic shift has resulted in a threefold increase in merged pull requests, accelerated release cycles, and a reduction in defects identified in later development phases.

Rashed Haq, GM's VP of autonomous vehicles, stated that software engineers previously spent only about 15% of their time writing code. The remaining 85% was dedicated to tasks such as analyzing vehicle data, triaging problems, running experiments, and testing potential fixes. AI agents are now employed to streamline these non-coding activities.

Haq emphasized that the substantial gains were achieved by fundamentally redesigning entire engineering workflows around AI agents, rather than simply adding coding assistants. He noted that providing only a coding chatbot does not address the inherent inefficiencies built into the broader development process.

GM's approach involved dividing the autonomous vehicle development into distinct loops: software development and testing in simulation, physical road testing, and post-deployment monitoring. The company identified and automated the longest bottleneck in each loop. AI agents were granted access to GM's internal tools and petabytes of data through customized servers and were guided by specific instruction documents for task execution.