The author has released Agent Control Plane v0.1, moving from theory to code to address governance in AI systems. This 'Kernel' release tackles the problem of unsafe, single-threaded agents in production environments. It introduces three core features designed to manage permissions, resources, and safety effectively for complex, multi-agent workflows.

Key capabilities include Async Support for non-blocking operations, allowing concurrent agent management without system lockups. ABAC (Attribute-Based Access Control) enforces logic-based rules, such as refund limits, rather than relying solely on role definitions. A Flight Recorder provides black-box audit logging for every Kernel decision, crucial for troubleshooting and compliance.

Beyond these features, the architecture supports a 'Shadow Mode' for safely testing agents against production inputs without executing tools. This allows developers to see what actions would have been taken, mitigating deployment risks. The release signals a shift from prompt engineering to robust policy engineering for enterprise-grade AI safety.