Pluggable Agentic AI Framework: A seven-layer composable enterprise agentic AI reference architecture on Amazon Bedrock AgentCore

This article presents the Pluggable Agentic AI Framework, a seven-layer composable reference architecture designed to move multi-agent AI applications from prototype to production scale on Amazon Bedrock AgentCore.

• Data and Knowledge layer provides relevant, up-to-date information through retrieval and embeddings
• Model and Inference layer uses configuration-driven model routing to switch between models based on task complexity
• Orchestration layer decomposes work into specialist agents with bounded autonomy and centralized discovery
• Security and Guardrails layer implements authentication, fine-grained access control, and data masking from day one
• Application layer keeps the user interface thin and delegates business logic to orchestration
• Observability layer provides end-to-end tracing, cost tracking, and dashboards across all layers
• Agentic Operations layer enables continuous evaluation for groundedness, accuracy, and safety
• Pluggability allows technology substitution at layer boundaries without redesigning the entire stack
• Reference implementation demonstrates a retail customer support scenario with supervisor and specialist agents

The framework emphasizes that production-ready agentic AI requires architecture-first design with security, governance, and observability built in from the start, not added afterward.