Get to your first working agent in minutes: Announcing new features in Amazon Bedrock AgentCore
New managed harness feature lets developers deploy working AI agents with just three API calls, no orchestration code.
AWS has announced significant updates to Amazon Bedrock AgentCore designed to eliminate the infrastructure barriers that have slowed AI agent development. The centerpiece is a new managed agent harness feature that replaces what previously required days of orchestration coding, framework wiring, and infrastructure setup. Instead of building custom backend plumbing from scratch, developers can now declare their agent's model, tools, and instructions through configuration and deploy with just three API calls. This allows teams to validate agent ideas in minutes rather than days, with support for popular frameworks including LangGraph, LlamaIndex, CrewAI, and AWS's own open-source Strands Agents.
The platform maintains flexibility alongside this simplified workflow. When projects require custom orchestration logic, specialized routing, or multi-agent coordination, developers can switch from configuration to code-defined harnesses using the same underlying platform with microVM isolation and persistent session state. AgentCore also introduces a unified CLI that keeps developers in one workflow across the entire lifecycle—from prototyping locally to deploying to production without switching tools or building separate pipelines. This addresses the common fragmentation where building and deploying agents required completely different processes and toolchains.
- Managed harness replaces days of infrastructure coding with configuration-driven deployment using just three API calls
- Supports major frameworks including LangChain, LlamaIndex, CrewAI and AWS's open-source Strands Agents
- Unified CLI enables prototyping, deployment, and operations from a single terminal workflow
Why It Matters
Removes the biggest bottleneck in agent development, letting teams focus on agent logic rather than infrastructure plumbing.