Blog: Amazon Bedrock AgentCore in practice with Apache SkyWalking AI Sessionizer (#942)

A LangGraph agent on Amazon Bedrock AgentCore Runtime, recorded by AI
Sessionizer while each session runs and kept after its execution
environment stops, with configuration rather than code: four LANGSMITH_*
variables for the conversation, three ASZ_* variables, asz-changes and the
LangChain shim in the image, and a shared Amazon EFS directory for the file
changes.

From a real run on 2026-10-10 in us-east-1 with Amazon Nova Pro: six talks
on one AgentCore session, 19 model calls and one recorded file change. The
post shows the history carried across invocations, what the tools changed
against what the agent said (an invented agentcore_idle_timeout), and a
turn sent on the same session after the 15-minute idle stop: a fresh
environment, a two-message prompt, a confident wrong recollection, while
the conversation in Sessionizer continued across the stop.

Figures: a hero and an idle timeline drawn with GPT-6 Astra, Figure 1
showing both recording paths, and screenshots of the replay page on Fargate
cropped for the post column. The demo is
https://github.com/wu-sheng/aws-agentcore-asz-demo.