The Three-Layer Agentic Architecture Most Teams Build Wrong
This article discusses the importance of a separate governance plane in agentic AI systems, beyond just the infrastructure and cognitive architecture layers.
Why it matters
Proper governance architecture is critical for scaling agentic AI systems and addressing security and risk concerns.
Key Points
- 1The two-layer framing of infrastructure and cognitive architecture is incomplete, missing the critical governance plane
- 2Embedding governance policies within the agent's code leads to brittleness, lack of audit evidence, and non-deterministic enforcement
- 3A separate governance plane can intercept and evaluate every agent action against defined policies, without modifying the agent code
- 4This allows for scalable governance updates across multiple agents without requiring code deployments
Details
The article explains that the widely-cited two-layer agentic architecture model (infrastructure and cognitive architecture) is missing a crucial third layer - the governance plane. The governance plane is responsible for enforcing policies on agent behavior at runtime, controlling what tools agents can access, what data they can touch, what outputs they can produce, and what cost they can incur, without living inside the agent itself. Unlike the cognitive architecture that determines what the agent tries to do, the governance plane determines what the agent is allowed to do, independent of its reasoning. This separation allows for policy updates to be made without modifying the agent code, enabling scalable governance across multiple agents. The article also discusses how embedding governance policies within the agent's code leads to brittleness, lack of audit evidence, and non-deterministic enforcement.
No comments yet
Be the first to comment