AWS makes Amazon Bedrock AgentCore Harness generally available at Summit New York

Amazon Web Services (AWS) made Amazon Bedrock AgentCore Harness generally available on June 17 at its summit in New York, opening a managed path for building and running production AI agents. The service lets developers declare an agent's model, tools, skills, and instructions in configuration and hands the orchestration loop to AWS, the company announced in a roundup of summit news.

The pitch is that teams can ship agents without writing and maintaining the control logic that runs them. AWS is positioning AgentCore as production infrastructure for agents rather than a prototyping tool, with the surrounding services aimed at the parts enterprises tend to struggle with: live data, private knowledge, and safety controls.

Alongside the release, AWS launched Web Search on Amazon Bedrock AgentCore, a fully managed tool that grounds agents in current web knowledge through a Model Context Protocol (MCP) connector on AgentCore Gateway. The company said the tool is available in the US East (N. Virginia) region at no additional cost.

AWS announced several related capabilities the same day. Amazon Bedrock Managed Knowledge Base provides an enterprise retrieval-augmented-generation pipeline with connectors to Amazon S3, SharePoint, Confluence, and Google Drive. Amazon Bedrock Guardrails now plug into AgentCore to evaluate agent actions for prompt injection, harmful content, and sensitive-data exposure. A separate feature, AWS Context, which builds a knowledge graph from existing data for agentic search, was described as coming soon rather than available now.

AWS said the number of tasks performed by agents on AgentCore grew 15-fold over the prior six months, a usage figure the company did not independently substantiate. The keynote was delivered by Swami Sivasubramanian, AWS vice president of agentic AI, at the Javits Center.

The announcements deepen AWS's competition with Microsoft, Google, and OpenAI to own the layer where companies build and govern AI agents. How well the managed orchestration and guardrails hold up under real production workloads will determine whether AgentCore becomes the default for enterprise agents or one option among several.