We present open-kraken, a control-plane architecture that treats large-scale AI agents as members of an organization rather than isolated chat participants. While individual agents have become remarkably capable, high-value tasks—enterprise operations, complex R&D, long-horizon planning, and robotic perception–planning–control loops—require hundreds of heterogeneous agents to coordinate reliably over days or weeks. The system introduces three minimal primitives: the Authoritative Execution Ledger (AEL), the Budget-Aware Cognitive Workload Scheduler (CWS), and Shared Execution Memory (SEM). On a 32-node cluster with 1,200 concurrent tasks, it achieves 94.2% success rate under 30% node failures and reduces cost by 31.4% via intelligent multi-provider routing. A logistics-network case study demonstrates its applicability to physical systems. The full open-source implementation is available at: https://github.com/open-kraken (or your actual repo link).This work shifts the focus from “making single agents smarter” to building the organizational infrastructure that lets them work together effectively at scale.