Agent Core provides production-ready components like runtime, memory, and observability so you can focus on agent logic instead of infrastructure.
Agent Core automatically syncs conversation events to short-term memory for session context and long-term memory for persistent knowledge.
The demo agent uses internal tools, an external server, and an AWS Lambda function via the Agent Core gateway to handle customer queries.
The AWS CloudWatch dashboard provides a high-level view of agent performance, including sessions, traces, token usage, and memory API calls.
A detailed walkthrough of a trace shows how the agent correctly calls two different tools to answer a complex user question about product price and warranty.
This example demonstrates a straightforward scenario where the agent correctly identifies and calls a single internal tool to retrieve a return policy.
When an agent fails to call the correct tool, the built-in correctness evaluation provides a low score and explains the factual error, helping you debug the agent's reasoning.
An agent can answer questions about its own capabilities directly from its instructions without needing to invoke any external tools.
Jobs that call for the skills explored in this talk.