AWS announces new capabilities for its AI agent builder

Amazon Web Services (AWS) is enhancing its AI agent platform, known as Amazon Bedrock AgentCore, to streamline the process of creating and overseeing AI agents for businesses. The company unveiled a range of new features for AgentCore during its annual AWS re:Invent conference on Tuesday. One significant addition is the Policy feature in AgentCore, which enables users to define interaction boundaries for AI agents using natural language. This feature works in conjunction with the AgentCore Gateway, ensuring that agents adhere to specified actions and halt any activities that breach established guidelines. For example, developers can configure an AI agent to automatically process refunds up to $100 while requiring human intervention for larger amounts, as explained by David Richardson, the vice president of AgentCore. In addition, AWS introduced AgentCore Evaluations, a toolkit comprising 13 pre-built evaluation systems designed to assess AI agent performance across various metrics, such as accuracy, safety, and tool selection. This suite not only aids developers in ensuring their agents meet essential standards but also facilitates the creation of custom evaluation features. "This will significantly alleviate the major concerns surrounding agent deployment," Richardson noted, emphasizing the value of these new capabilities. Another exciting development is the introduction of AgentCore Memory, which empowers agents to accumulate user information over time—like flight preferences or hotel choices—and utilize this data for future interactions. "These enhancements represent our ongoing commitment to refining the different layers of AgentCore," Richardson stated, highlighting the integration of Policy, the empowerment of agents through Memory, and supporting developer efforts. While the future of AI agents is a topic of debate, with some skeptics questioning their longevity, Richardson is optimistic about the resilience of the tools being developed. He believes that the ability to leverage the reasoning capabilities of AI models in conjunction with practical applications will lead to sustainable innovation. "Although the landscape may evolve, we are confident in our readiness to adapt to emerging trends," he concluded.