Amazon Bedrock AgentCore harness reaches general availability with a two-call deploy model

AWS Machine Learning Blog·medium signal

AWS made the managed AgentCore harness generally available (June 18), letting teams stand up production agents with just CreateHarness (define model, tools, skills, memory) and InvokeHarness (run it). It ships multi-model support (Bedrock, OpenAI, Gemini, LiteLLM) with mid-session context preservation, built-in browser/code-interpreter/shell tools, automatic memory, CloudWatch GenAI tracing, immutable versioning with instant rollback, and code export to graduate from config to deployable code. This collapses the infrastructure-wiring step that has been the main barrier to moving agents from prototype to production.