postmortem.io Beta
Shutdown

Amazon Bedrock Agents Classic maintenance mode

AWS Read the original company blog ↗

Amazon Bedrock Agents Classic maintenance mode

AWS’s wind-down notice for Amazon Bedrock Agents — its first-generation agent product, launched November 2023 — announced in the June 30, 2026 service availability update and superseded by Amazon Bedrock AgentCore.

Amazon Bedrock Agents (launched November 2023) is now Amazon Bedrock Agents Classic and will no longer be open to new customers starting on July 30, 2026. For capabilities similar to Amazon Bedrock Agents Classic, explore Amazon Bedrock AgentCore. Amazon Bedrock continues to be available, but the Agents Classic service will be in maintenance mode.

This guide provides information about the maintenance mode migration options for Amazon Bedrock Agents Classic customers.

Note: Amazon Bedrock continues to be fully supported. Only the Amazon Bedrock Agents Classic service is no longer open to new customers as of July 30, 2026. Your existing Amazon Bedrock models, Knowledge Bases, and Guardrails are not affected.

Migration plan

We recommend that you migrate your Amazon Bedrock Agents Classic workloads to Amazon Bedrock AgentCore. AgentCore is the platform to build, connect, and optimize AI agents. The managed harness in AgentCore provides a config-based starting point where developers can declare the agent’s model, tools, and instructions. AgentCore handles the environment, compute, memory, identity, and observability.

The managed agent harness in AgentCore provides a config-based starting point where builders can declare the agent’s model, tools, and instructions. AgentCore handles compute, environment, memory, identity, and observability. The harness supports:

  • Managed orchestration loop with built-in tool connectivity
  • Action groups exposed as MCP tools via AgentCore gateway (REST APIs, Lambda functions, or code-level @tools)
  • Gateway-fronted knowledge base integration
  • Inline function tools for return-of-control and human-in-the-loop patterns
  • Code interpreter for sandboxed code execution
  • Short-term and long-term memory with configurable strategies
  • Guardrail enforcement via AgentCore gateway
  • Persistent, end-to-end tracing of all agent actions