GM Tripled Merged Pull Requests After Rebuilding Engineering Workflows Around AI Agents

Quick Facts

• GM’s autonomous vehicle engineering team tripled its merged pull requests after redesigning workflows around AI agents, not just adding coding tools.
• CEO Mary Barra said nearly 90% of code written by GM’s autonomy team is now AI-generated, stress-tested in simulations running roughly 100 years of human driving per day.
• GM laid off more than 600 IT employees — over 10% of the department — to make room for workers with AI-focused backgrounds.

General Motors rebuilt its autonomous vehicle engineering operation around AI agents and tripled the number of merged pull requests, Rashed Haq, GM’s VP of Autonomous Vehicles, said at VB Transform 2026 in Menlo Park, California.

The gains came from workflow redesign, not tool adoption. Haq said GM identified the longest bottleneck in each development loop, automated it, then repeated the process across simulation testing, public road testing, and post-deployment monitoring.

“If you give somebody just a chatbot which can do coding, there’s still a lot of inefficiency built into that process,” Haq said during the main-stage fireside chat. The increase in merged pull requests also produced faster feature releases and fewer bug escapes, he said.

What Engineers Actually Do

GM’s engineers spend only 15% of their time writing code. The remaining 85% covers analyzing vehicle data, triaging problems, running experiments, and testing fixes. AI agents now handle large portions of that work.

The 15% figure is not new. A 2019 Microsoft study of nearly 6,000 professional developers found they spent roughly 14% to 20% of an eight-hour workday writing code. GM’s agents are attacking the time that was already consuming most of an engineer’s day.

During GM’s Q1 2026 earnings call, CEO Mary Barra said nearly 90% of code produced by the autonomy team is AI-generated. The company runs simulations capable of replicating roughly 100 years of human driving every day to stress test that output.

The Technical Architecture

GM connected agents to internal tools and company data through customized Model Context Protocol servers. It also created version-controlled instruction documents, called “skills,” that tell agents how to complete specific tasks.

One high-value application automates vehicle telemetry. Agents analyze data from public road vehicles, conduct an initial triage, and create issues for engineers to investigate. They can also call the underlying tools used by WebViz, GM’s telemetry visualization system, bypassing the graphical interface entirely.

A previous computational task that required 15 hours now completes in under one minute using an AI-based probabilistic method.

Human oversight is built into the system. Agents propose fixes, run tests, open pull requests, and annotate changes. Engineers retain final approval and review performance measurements before work advances to production.

Four Engineers Changed the Whole Organization

GM assigned four engineers to an internal agent platform team and deployed them directly inside engineering groups. Their job was to identify useful workflows, document what worked, and spread successful practices across teams.

Sterling Anderson, GM’s chief product officer, called it “a new operating system for product development.” Anderson previously led Tesla’s Autopilot development and co-founded autonomous trucking company Aurora Innovation before GM hired him last June for a reported $40 million package.

Anderson said the industry average for developing a new vehicle is four to five years. GM’s target is two. “When you can cut that period more than in half, you’re getting just better vehicles evolutionarily,” he said.

Workforce Shift

The efficiency gains have a human cost. GM laid off more than 600 salaried IT employees, representing over 10% of its IT department, in a deliberate skills swap. The company confirmed the cuts to TechCrunch after Bloomberg News first reported them.

GM said it is now hiring for workers who can build AI systems from the ground up, designing the architecture, training models, and engineering data pipelines, not workers who use AI as a productivity add-on.

Read more: GM redesigned its engineering workflows around AI agents — and tripled its merged pull requests