Introduction to AgentCore

Foundation models have become increasingly capable at planning steps, generating text, and following complex instructions. Yet the gap between a working prototype and a production AI agent remains significant, and it is often less about the model itself and more about the surrounding system. Many of the hard problems are infrastructure and operations problems. An agent that generates Python code needs an isolated sandbox to run it safely. An agent that assists customers across multiple sessions needs persistent memory or state that carries across sessions. An agent that files Jira tickets or updates Salesforce records needs secure credential management to prevent OAuth tokens from being stored or exposed in application code. When multiple agents across teams need access to enterprise tools, the platform needs governance over which agents can call which APIs.

Without managed solutions for these challenges, engineering teams spend months building custom plumbing, including container orchestration for safe code execution, token vaults for credential storage, and session stores for memory persistence, before a single agent reaches production. This operational overhead delays time-to-value and introduces security risks that compound at scale.

Amazon Bedrock AgentCore addresses these gaps directly. It is a collection of fully managed infrastructure services purpose-built for production AI agent deployments. Rather than replacing agent orchestration frameworks, AgentCore provides the foundational layer beneath them, handling the infrastructure so teams can focus entirely on agent logic.

Note: AgentCore is framework-agnostic. It works with native Bedrock Agents, LangChain, CrewAI, and custom orchestration code alike.

By the end of this lesson, you will be able to identify the four AgentCore services and their distinct roles, explain the Model Context Protocol (MCP) and how AgentCore Gateway implements it, differentiate AgentCore from native Bedrock Agents, and map real-world business scenarios to the appropriate AgentCore components. The following section breaks down each service in the AgentCore family.

The AgentCore service family

AgentCore is not a single monolithic service. It is a family of four distinct, composable services, each targeting a specific production infrastructure gap that agents encounter when moving from prototype to deployment. Think of it like a toolkit: each tool solves a well-defined problem, and you pick the combination your agent needs.

The four services

Each service operates independently and can be composed with others depending on the ...