AWS is expanding Amazon Bedrock AgentCore with a managed agent harness, a new CLI, and prebuilt coding skills. Developers can now set up a working AI agent with just 3 API calls, without writing orchestration code themselves.
Anyone looking to build an AI agent typically has to spend days setting up infrastructure before the agent can perform even a single task. Compute, sandboxing, authentication, and persistent storage—teams have to piece those building blocks together themselves. Only then could they determine whether the agent was even useful. AWS aims to break that pattern with new functionality in AgentCore, the platform within Amazon Bedrock for building and deploying AI agents.
Tip: AWS is stepping on the AI gas pedal for greater depth and broader adoption
Managed harness: from idea to working agent in minutes
The new managed agent harness replaces manual agent infrastructure setup with configuration via three API calls. Developers define which model the agent uses, which tools are available, and which instructions apply. AgentCore then sets up compute, tooling, memory, identity, and security for you. Trying a different model or adding a tool is a configuration change, not a code rewrite.
The harness runs on Strands Agents, AWS’s open-source framework, and supports popular frameworks such as LangGraph, LlamaIndex, and CrewAI. For teams seeking greater control, AgentCore offers the option to switch from a config-based to a code-defined harness, on the same platform, with the same microVM isolation. Session state is stored on a persistent filesystem, allowing agents to pause a task in the middle and resume later.
“Previously, prototyping each new agent required days of orchestration code and infrastructure setup before we could validate an idea,” says Rodrigo Moreira, VP of Engineering at VTEX. “We can now validate agent ideas in minutes instead of days.”
CLI and coding skills round out the picture
In addition to the harness, AWS is introducing a new AgentCore CLI that allows developers to manage the entire lifecycle, from prototype to production, from a single terminal. Deployments are handled via infrastructure-as-code with CDK support. Terraform integration will follow later. What is tested locally runs identically in production.
Finally, AWS is introducing pre-built coding skills for AgentCore. These provide coding assistants Claude Code, Codex, and Cursor with up-to-date knowledge of AgentCore best practices, ensuring recommendations align with how the platform is intended to be used—not just which API endpoints exist. Kiro already includes these skills as a built-in Power.
The managed agent harness is available today in preview in US West (Oregon), US East (Northern Virginia), Asia Pacific (Sydney), and Europe (Frankfurt). The CLI and the persistent agent file system are available in all commercial AWS regions where AgentCore is offered. Coding skills will follow by the end of April. There are no additional costs for the CLI, harness, or skills. Users pay only for the resources they use.