Amazon Bedrock AgentCore Adds Managed Harness and CLI Features
Developers can now launch their first functional AI agent in minutes thanks to new features in Amazon Bedrock AgentCore. These updates cut out the usual infrastructure hurdles that delay testing agent performance. Teams often spend days on frameworks, storage, authentication, and deployment before checking if the agent works well. AgentCore changes that by letting builders concentrate on agent behavior using familiar tools like LangGraph, LlamaIndex, CrewAI, and Strands Agents.
Quick Path from Concept to Active Agent
Every AI agent needs an orchestration layer to manage model calls, tool selection, result handling, context, and error recovery. This requires compute resources, secure code execution environments, tool connections, storage, and recovery systems, which together form the agent harness. Previously, teams built this harness manually each time, picking frameworks, coding orchestration, linking tools and memory, and handling security before any testing.
The new managed agent harness in AgentCore simplifies this to three API calls without custom orchestration code. Developers specify the model, available tools, and instructions. AgentCore then assembles compute, tools, memory, identity, and security for an instantly testable agent. Switching models or tools means just updating configuration, not rewriting code. Multiple agent versions can run tests in minutes via parameter changes.
Flexibility remains intact. The harness uses Strands Agents, an open source framework from AWS. For custom logic, routing, or multi-agent setups, users shift to code-based harnesses on the same platform with microVM isolation and deployment processes. AgentCore stores session state on a durable filesystem, allowing task suspension and resumption. This supports human-in-the-loop workflows without extra setup or later redesigns. Start simple in minutes and scale as needs grow, no rearchitecture required.
"We're building AI agents that will revolutionize ecommerce," said Rodrigo Moreira, VP of Engineering at VTEX. "Previously, prototyping each new agent required days of orchestration code and infrastructure setup before we could validate an idea. The harness feature in AgentCore will change that: swapping a model, adding a tool, or refining an agent's instructions is now a configuration change, not a rebuild. We can now validate agent ideas in minutes instead of days, and we're looking forward to accelerating agent development further with these new capabilities."
Single Terminal for Build, Deploy, and Operations
Once an agent works locally, production often involves new tools, pipelines, and environments unrelated to the build process. The AgentCore CLI unifies the entire lifecycle in one terminal: prototyping, deployment, and operations. Iterate locally, then deploy without tool switches or custom pipelines. Deployments use infrastructure as code via CDK support, with Terraform support coming soon. Configurations stay reproducible and version-controlled, matching local tests to production exactly.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Better Context for Coding Assistants
Developers often pair agent work with coding helpers like Claude Code or Kiro. These tools depend on good context for effectiveness. While general servers provide API and documentation access, they lack guidance on best patterns, capability integration, or common task paths. AgentCore's new pre-built skills deliver curated, up-to-date knowledge of platform practices. Suggestions align with intended usage, not just endpoints. Kiro integrates this now as a built-in Power, with plugins for Claude Code, Codex, and Cursor arriving soon. On a fast-changing platform, this reduces errors from the start.
Launch Details and Next Steps
The managed agent harness appears in preview today in US West (Oregon), US East (N. Virginia), Asia Pacific (Sydney), and Europe (Frankfurt). AgentCore CLI and persistent agent filesystem work in all AWS commercial regions offering AgentCore. Coding agent skills arrive by month's end in April. Costs cover only used resources, with no extras for CLI, harness, or skills. Check AgentCore documentation to begin.
These tools keep focus on agent logic, free from infrastructure worries. Add evaluations, memory, tools, and policies as agents advance, all on the same platform from prototype to production.
Amazon Bedrock, part of AWS services, provides access to foundation models from leading AI companies for building generative applications. AgentCore extends this for agent development, handling the underlying complexities.
VTEX operates as a digital commerce platform serving global enterprises, where AI agents enhance operations.