At the AWS re:Invent 2025 conference, Amazon unveiled significant enhancements to its Bedrock AgentCore platform, designed for building and deploying AI agents. A key feature, “Policy in AgentCore,” empowers development teams to set operational boundaries for agents using natural language, enabling clearer guidelines on tool access and actions. The update also introduces AgentCore Evaluations, which aids software engineering teams in assessing agent performance in real-world scenarios, complete with 13 pre-built evaluators. Additionally, the new AgentCore Memory feature allows agents to learn from past interactions, enhancing decision-making through episodic functionality. This capability not only improves operational efficiency but also reinforces security by mitigating unauthorized data access. AWS emphasizes the importance of establishing robust controls to prevent mistakes, ensuring that while agents operate autonomously, they do so within defined parameters. Overall, these advancements highlight AWS’s commitment to making AI agents both powerful and secure for businesses.
Source link
