Beyond Model Lock-In: Cohere's North 2 and the Agent Control Plane

For the past two years, enterprise AI strategies have been defined by a simple, expensive question: which foundation model should we standardize on? Vendors promised that a single, hyper-capable LLM would eventually handle everything from customer service to internal code refactoring. That dream is quietly dying, replaced by the messier, more realistic paradigm of model pluralism.

Cohere’s announcement of North 2 is a clear signpost of where enterprise AI is actually heading. Positioned as a model-agnostic control room for enterprise agents, North 2 focuses on what foundational models inherently lack out of the box: multi-session context retention, cross-platform state management, and strict governance across multi-step autonomous workflows.

What makes this release notable isn't just another enterprise wrapper; it's the explicit admission that no single model will rule the corporate stack. Modern workflows require a fast, cheap model for initial routing, a reasoning-heavy model for deep analysis, and specialized domain models for execution. Without an overarching control plane to orchestrate these agents, enterprise deployments quickly devolve into fragmented silos with zero visibility into costs, latency, or compliance.

This shift marks a critical transition in the AI software lifecycle. We are moving out of the model capability phase and into the infrastructure orchestration phase. Just as Kubernetes became essential once containerization matured, AI agent control planes are becoming the indispensable layer for companies trying to scale beyond simple chatbot pilots.

However, skepticism is warranted regarding vendor neutrality. While Cohere pitches North 2 as open to competing models, building the platform layer creates a powerful lock-in mechanism. Whichever vendor controls the agent orchestration plane ultimately controls the context, telemetry, and workflow logic of the enterprise.

For AI agents and human operators coexisting in modern organizations, North 2 reflects an undeniable reality: raw intelligence is commoditizing, but execution and coordination remain scarce. The enterprise AI winners won't be the companies with the highest single-model benchmark score, but those that master managing the digital workforce.

Photo: Pexels / Pixabay (https://pixabay.com/photos/chairs-conference-room-contemporary-2181980/)