Amazon Bedrock AgentCore harness is now generally available: Go from idea to production-grade agent in minutes
TL;DR
AWS made Amazon Bedrock AgentCore harness generally available on June 18, 2026. AWS says teams can define an agent with CreateHarness and run it with InvokeHarness, without writing orchestration code or building a container by default. The harness runs each agent in an isolated environment with a filesystem and shell. It can read and write files, run commands, load skills, use browser and code interpreter tools, and connect to tools through AgentCore Gateway, remote MCP, or inline functions.
Nauti's Take
The hardest part of agents lives in sandboxing, logs, memory, and tool permissions. AWS is turning that grey zone into a managed product.
Builders get less Docker yoga, but the bar moves to real evals, clean permissions, and runtime control that survives production traffic.
Briefingshow
Enterprise agents rarely fail only because of the model. The hard work is isolation, auth, tool wiring, memory, logs, cost controls, evaluations, and rollback. AWS is turning that operational layer into configuration.
For teams already on AWS, this can shorten prototypes, but it also pulls agent operations deeper into Bedrock, CloudWatch, IAM, and AgentCore primitives.
Sources
