GM Triples Code Merges by Redesigning Workflows Around AI Agents
General Motors redesigned its autonomous vehicle engineering workflows around AI agents rather than simply adding coding assistants, resulting in a tripling of merged pull requests and faster releases with fewer defects. GM's VP of autonomous vehicles noted that engineers spend only 15% of their time writing code, with agents now handling analysis, triage, experimentation, and testing across the remaining 85% of development work. The company connected agents to internal tools and petabytes of data through customized Model Context Protocol servers and version-controlled instruction documents.
TL;DR
- GM engineers spend only 15% of time writing code, with AI agents handling the other 85% including data analysis, problem triage, experiments, and testing
- GM tripled merged pull requests across its autonomous vehicle organization by redesigning complete workflows around agents, not just adding coding assistants
- Agents access internal tools and vehicle telemetry data through customized MCP servers and can run machine-learning experiments in parallel
- Agent permissions mirror engineer permissions, with engineers remaining accountable for agent output and all findings presented in human-readable format
Why It Matters
This demonstrates a fundamental shift in how AI augments engineering work. Rather than automating coding itself, GM automated the surrounding workflow bottlenecks that consume most developer time. The approach validates earlier research showing developers spend only 14-20% of their time actually writing code, suggesting that workflow redesign around AI agents may yield larger productivity gains than coding assistants alone.
Business Impact
Tripling merged pull requests while reducing escaped defects directly improves development velocity and product quality in a safety-critical domain. For autonomous vehicles, faster iteration cycles and fewer defects reaching production stages have significant implications for time-to-market and liability exposure. The model also shows how enterprise AI adoption requires infrastructure investment and organizational change beyond tool procurement.
Key Implications
- Agentic AI's value in engineering may lie more in workflow orchestration and bottleneck elimination than in code generation itself
- Enterprise deployment of AI agents requires custom infrastructure, permission models, and accountability frameworks rather than off-the-shelf solutions
- Identifying and automating the longest bottleneck in each development loop, then repeating the process, may be a replicable methodology for other organizations
What to Watch
Monitor whether GM's three-fold increase in merged pull requests translates to faster autonomous vehicle releases and improved safety metrics in production. Track whether other automotive or software-heavy organizations adopt similar workflow-redesign approaches and what infrastructure investments they require. Watch for emerging best practices around agent permission models and accountability frameworks as enterprise AI agent deployment scales.
Subscribe to the newsletter
The latest stories and analysis, delivered to your inbox.
Free. No spam. Unsubscribe any time.
