Accelerate Your AI Agent Development with AgentCore
Seamlessly Transition from Idea to Working Agent in Three Steps
Build, Deploy, and Operate Your Agents from a Unified Workflow
Equip Your Coding Agents with Essential Context
Get Started with AgentCore Today
About the Authors
Introducing AgentCore: Streamlining Agent Development from the Ground Up
Getting an AI agent up and running has traditionally involved tackling a myriad of infrastructure challenges before you can even begin to evaluate the agent’s performance. Developers often find themselves entangled in the complexities of frameworks, storage solutions, authentication, and deployment pipelines. By the time the agent can take on its first task, invaluable time has often been consumed, diverting focus from what truly matters—building agent logic.
Unveiling AgentCore
To address these enduring inefficiencies, we created AgentCore from the ground up, enabling developers to concentrate on the core logic of their agents rather than getting bogged down by backend configurations. AgentCore is designed to integrate seamlessly with popular frameworks and models, including LangGraph, LlamaIndex, CrewAI, and Strands Agents. Today, we are excited to announce new capabilities that further simplify the agent building experience, effectively eliminating infrastructure barriers at every stage of development—from prototyping to production deployment.
Go from Idea to Running Agent in Three Simple Steps
Every agent requires an orchestration layer, a critical component that governs the model’s operations, manages tools, and handles errors. Traditionally, establishing this layer involved a significant amount of foundational work: selecting frameworks, writing orchestration code, integrating tools, and ensuring security and storage solutions—all before the agent could even begin processing requests. Our research indicates that most teams expend days wrestling with infrastructure, delaying the testing of their agent’s real-world utility.
Enter AgentCore’s Managed Agent Harness
With our new managed agent harness feature, we replace this tedious setup with a straightforward configuration process. You can launch your agent with just three API calls—no orchestration code required. Simply declare your agent, specify which model it will utilize, the tools it will interact with, and the instructions it will follow. AgentCore’s harness seamlessly combines compute resources, tooling, memory, identity, and security, allowing you to test a fully operational agent in mere minutes. If you want to try a different model or add a tool, it’s just a configuration modification, not a complete code overhaul.
Speed and Flexibility Hand in Hand
Despite the substantial speed offered by AgentCore, flexibility remains uncompromised. Powered by Strands Agents—an open-source platform from AWS—users can switch from a config-based harness to a code-defined harness when custom orchestration logic or advanced multi-agent coordination is needed. AgentCore even retains session states to a durable filesystem, enabling agents to pause mid-task and resume seamlessly, making human-in-the-loop processes straightforward and eliminating the need for complex redesigns later on.
“We’re building AI agents that will revolutionize eCommerce,” stated Rodrigo Moreira, VP of Engineering at VTEX. “Previously, developing each new agent required days of configuration and orchestration code setup before we could validate our ideas. The harness feature in AgentCore simplifies this process: changing models, adding tools, or refining instructions is now just a matter of configuration, not a complete rebuild. We can now validate our agent concepts in minutes instead of days, significantly speeding up our development efforts.”
A Unified Workflow from Development to Deployment
Once your agent is primed for production, transitioning to a deployment phase often involves a frustrating pivot away from your development environment. You usually have to set up a new deployment pipeline, configure environments, and integrate numerous tools.
Introducing the AgentCore CLI
The newly launched AgentCore CLI allows you to maintain a consistent workflow across the entire lifecycle of your agent—from prototyping to deployment to operation—right from your terminal. This innovation enables you to iterate on your agent locally and deploy it without needing to switch tools or build a separate processing pipeline. AgentCore supports infrastructure as code (IaC) through CDK and Terraform (coming soon), ensuring that your agent’s configuration is consistent, reproducible, and version-controlled, meaning that what you tested locally mirrors what runs in production.
Enhancing Context for Coding Agents
Many developers rely on coding assistants like Claude Code or Kiro during the development journey. However, the effectiveness of these tools is heavily contingent on the context they possess. A generic MCP server might provide access to APIs and documentation, but it fails to encapsulate important nuances: the right patterns to follow, how different capabilities align, and optimal paths for common tasks.
New Pre-Built Skills in AgentCore
Our platform introduces pre-built skills that go beyond basic API access. These skills endow coding agents with a well-curated understanding of AgentCore best practices, ensuring that the suggestions and guidance provided are not merely theoretical but reflect the platform’s intended usage. Kiro has integrated these features, with Plugin support for Claude Code, Codex, and Cursor coming soon.
Get Started with AgentCore
The managed agent harness in AgentCore is now available in preview across four AWS Regions: US West (Oregon), US East (N. Virginia), Asia Pacific (Sydney), and Europe (Frankfurt). The AgentCore CLI and persistent agent filesystem are accessible in all AWS commercial Regions where AgentCore is offered. The coding agent skills are set to launch by the end of April. You only pay for the resources you use, with no additional charges for the CLI, harness, or skills (learn more on the AgentCore pricing page). Visit the AgentCore Documentation to dive in.
With these robust capabilities, developers can focus on creating sophisticated agent logic, free from the shackles of infrastructure concerns. As your agent evolves, adding memory, tool connections, and policy enforcement is as simple as a configuration update—without requiring a complete rearchitecture. The platform you prototype on is the same as the one you deploy, ensuring consistency throughout.
About the Authors
Madhu Parthasarathy, General Manager of Amazon Bedrock AgentCore, holds over 20 years of expertise in large-scale distributed infrastructure. After more than 16 years at Amazon, where he led several initiatives, he now spearheads the development of AgentCore. Madhu has also held influential positions at LinkedIn and other companies, emphasizing security and developer experience in AI infrastructure. He is currently based in Santa Clara, California.
As we advance into an era where AI agents will play transformative roles in various sectors, let’s embrace the efficiency and agility that platforms like AgentCore offer to nurture innovation and creativity.