How the BMW Group governs hundreds of agents across thousands of developers
How the BMW Group governs hundreds of agents across thousands of developers
BMW Group R&D describes how it governs agentic AI for more than 12,500 developers, using a framework-agnostic agent harness, externalized governance at an MCP gateway and central registry, and graduated per-step autonomy levels from human-in-the-loop to autonomous within guardrails. The approach codifies five practices as infrastructure as code, including CI/CD verification and evaluation against ground-truth datasets, a three-tier composition hierarchy of MCP tools, skills and flows, and a registry that serves as discovery, governance, security and deduplication checkpoint. A governance agent supports audits with human confirmation, and runtime observability catches data and model drift.
Source: AWS Machine Learning Blog · aws.amazon.com