Breaking02 / IT
AWS details seven-layer reference architecture for AI agents
The framework lets teams swap the technology within each layer, but the announcement does not name a licence for its sample repository.

AWS on 10 Nov 2026 published a seven-layer reference architecture for AI agents that runs on Amazon Bedrock AgentCore, and its announcement names no licence for the accompanying sample repository.
Scope. The architecture is called the Pluggable Agentic AI Framework, and a workshop and a public sample repository accompany it, according to the AWS blog. It is not a new model, so it has no weights or VRAM requirements.
The first three layers are data and knowledge, model and inference, and orchestration, which divides up the work. The others are security and guardrails, application, observability and agent operations. AWS said the services shown in each layer are not mandatory choices.
AWS said the contracts between layers are fixed, so the underlying technology in a layer can be replaced on its own. It also described using a proxy such as LiteLLM to switch models without changing application code.
Sample implementation. The AWS example is a retail customer support scenario covering order status and refund eligibility. A supervisor agent takes each request and hands it to specialist order and refund agents. Tool calls pass through access controls in AgentCore Gateway first.
AWS wrote that most agent projects stall before reaching production because each team rebuilds its own routing, guardrails and observability. It added that the sample implementation does not replace production security hardening, testing and review.
The announcement does not name the repository's licence or say how far it may be redistributed or modified.