InfoQ proposes a 'context as control plane' agent architecture

At 21:00 on 12 November 2026, InfoQ published an article proposing a new agent architecture paradigm described as "context as the control plane". The piece argues that context should function as a centralized governance layer spanning tools, sessions, and systems, rather than being hard-coded into individual applications, AI models, or microservices. It observes that as the industry moves from chatbots toward autonomous agents, the architectural bottleneck is shifting away from data retrieval toward the coordinated management of memory and knowledge, with the central challenge becoming determining what an agent should know, when it should call on what it already knows, how long it should retain information, and what it should forget.

The proposed design separates system knowledge from memory. System knowledge is framed as the system's official textbook, carrying policies, governance rules, and hard facts while preserving core business logic and database content. Memory is described as the key layer connecting historical data with real-time signals and translating them into an inferable state, deciding per task what needs to be remembered. The article also introduces a context engine that handles selection, filtering, and compression before information enters the model, and compares context to the immediate, active information sitting directly inside the model's active context window, much like RAM in a computer. It further notes that supplying incorrect context can significantly raise the chance of model hallucination. As of now, the paradigm remains a conceptual and architectural proposal set out in the InfoQ piece.