How GM Transformed Engineering Workflows with AI Agents

The automotive industry is undergoing a seismic shift, not just in the vehicles it produces but also in the workflows that drive innovation. General Motors (GM), a stalwart of American automotive engineering, is leading this charge by embracing artificial intelligence (AI) to reshape its engineering processes. According to Rashed Haq, Vice President of Autonomous Vehicles at GM, software engineers in the company now spend only 15% of their time writing code, with AI agents accelerating the remaining 85%. This radical approach has resulted in a staggering threefold increase in merged pull requests and a significant reduction in development defects.

At the recent VB Transform 2026 conference, Haq shared insights into how GM's engineering workflows have been meticulously redesigned around AI agents, rather than simply adding a coding assistant to existing processes. This transformation is not just a technological upgrade; it represents a fundamental rethinking of how software development can be streamlined, making it a case study worth exploring for companies across various sectors.

Understanding the Shift: From Coding to Comprehensive Workflows

The realization that software engineering encompasses much more than just writing code is not new. Research shows that developers typically spend less than a quarter of their time actually coding. A 2019 Microsoft study highlighted that developers averaged 96 minutes of coding on good days and 66 minutes on bad ones. Similarly, a 2018 survey from Stripe indicated that developers spend upwards of 17 hours a week on maintenance tasks like debugging and refactoring. These statistics underscore Haq's assertion that coding is merely one facet of the multifaceted engineering process.

GM's approach involves breaking down their autonomous vehicle development into several iterative loops: simulation, road testing, and post-deployment monitoring. By identifying bottlenecks in these loops, GM has been able to apply AI to automate the most time-consuming tasks and continually optimize their workflows. The focus is on creating a more efficient cycle that continually feeds improvements back into the system.

AI Agents: The Backbone of GM's Transformation

The AI agents at GM are not mere coding assistants; they are integral components of the engineering ecosystem. These agents gain access to extensive internal tools and a wealth of data through a customized Model Context Protocol (MCP). This enables them to conduct a range of tasks, from analyzing telemetry data collected from vehicles on public roads to triaging issues that require human attention.

How AI Agents Work

AI agents are designed to enhance human capabilities rather than replace them. They operate under a set of predefined instructions, or “skills,” that guide them through specific tasks. For instance, when analyzing vehicle telemetry, an AI agent can:

• Identify potential problems based on historical data
• Locate affected components in the vehicle
• Generate human-readable reports summarizing findings

This level of automation allows engineers to focus on higher-level problem-solving rather than getting bogged down in data analysis and troubleshooting. Furthermore, permissions assigned to these agents align closely with the engineers' access, ensuring accountability and oversight remain intact.

Real-World Impact: Increased Efficiency and Quality

The results of GM's integration of AI agents into their engineering workflows are significant. The company has reported a threefold increase in merged pull requests, indicating not just greater code volume but an improved velocity in releasing new features. Additionally, the increase in efficiency has led to fewer bugs escaping into later stages of development.

Haq notes that the internal agent platform was treated as a product, with dedicated engineers working alongside teams to identify valuable workflows, share best practices, and facilitate the adoption of AI tools. This collaborative approach has led to rapid iterations of the software development cycle, with engineers now able to run machine-learning experiments concurrently to test various hypotheses without waiting for lengthy manual processes.

Maintaining Quality Control

Despite the increased automation, quality control remains a priority at GM. Structured and unstructured tests are established at critical control points, allowing engineers to review performance measurements and ensure that each test aligns with its intended goals. This oversight is crucial in the autonomous vehicle space, where safety and reliability are paramount.

The Broader Implications for the Automotive Industry and Beyond

GM's successful integration of AI agents into its engineering workflows is not just a win for the company; it sets a precedent for the broader automotive industry and other sectors reliant on complex engineering processes. As companies strive for greater efficiency and innovation, the lessons learned from GM's experience can serve as a blueprint for similar transformations.

For other organizations looking to implement AI-driven workflows, the key takeaways from GM’s approach include:

• **Start with a full understanding of your workflows**: Map out every step of the process from problem identification to solution implementation.
• **Prioritize automation of bottlenecks**: Identify the most time-consuming tasks and focus AI efforts on those areas.
• **Maintain human oversight**: Ensure that accountability remains with engineers who must verify AI outputs and maintain quality control.
• **Treat AI integration as a collaborative product**: Designate teams to work directly with engineers to identify useful applications and facilitate adoption.

Key Takeaways

• GM has tripled its merged pull requests by redesigning workflows around AI agents.
• AI agents automate the majority of tasks outside of coding, improving efficiency and reducing defects.
• Quality control remains critical, with engineers maintaining oversight of AI-generated outputs.
• The automotive industry's shift towards AI transformation can serve as a model for other sectors.

Frequently Asked Questions

What are AI agents, and how do they differ from traditional coding assistants?

AI agents are sophisticated software programs designed to perform a variety of tasks autonomously or semi-autonomously within a defined workflow. Unlike traditional coding assistants that primarily focus on writing code, AI agents analyze data, triage problems, and execute tasks to optimize the entire development process. This capability allows human engineers to concentrate on more complex problem-solving tasks.

How has GM's approach impacted its engineering teams?

GM's integration of AI agents has significantly enhanced the productivity of its engineering teams. By automating repetitive tasks and streamlining workflows, engineers can now dedicate more time to coding and high-level problem-solving. The result is faster feature releases and a reduction in defects, leading to higher quality software for autonomous vehicles.

Can other industries apply GM’s AI integration strategies?

Absolutely. GM’s strategies for integrating AI into engineering workflows can be adapted to various industries, especially those focused on complex software development, such as finance, healthcare, and technology. By mapping workflows, identifying bottlenecks, and maintaining human oversight, organizations can leverage AI to drive efficiency and innovation.

What challenges might companies face when implementing AI in their workflows?

Challenges may include resistance to change among staff, the need for training in new technologies, and ensuring data quality for AI algorithms. Additionally, maintaining accountability and oversight in automated processes is crucial to prevent errors and ensure compliance with industry standards. Companies must address these challenges through effective change management and continuous feedback loops.