Amazon Web Services is enhancing its AI agent platform, Amazon Bedrock AgentCore, to simplify the process of building and monitoring AI agents for enterprises. The company announced several new features for AgentCore during its annual AWS re:Invent conference.
One significant upgrade is the introduction of Policy in AgentCore. This feature allows users to set boundaries for agent interactions using natural language. These boundaries integrate with the AgentCore Gateway, which connects AI agents with outside tools, to automatically check each agent’s actions and halt those that violate the established controls. Developers can set access limits for internal data or third-party applications like Salesforce or Slack. These policies can also instruct an AI agent to automatically issue refunds up to a certain amount, such as one hundred dollars, while requiring human approval for larger sums.
AWS also announced AgentCore Evaluations, a suite of thirteen pre-built evaluation systems for AI agents. These systems monitor factors including correctness, safety, and tool selection accuracy. This provides developers a foundation for building their own custom evaluation features and addresses common concerns about deploying agents.
Additionally, AWS is building a memory capability into the platform called AgentCore Memory. This feature allows agents to develop a log of information about users over time, such as travel preferences, and use that data to inform future decisions and interactions.
These three areas represent continued iteration on the AgentCore platform, focusing on connecting with existing systems through Policy, enhancing agent capability with Memory, and supporting development teams with Evaluations.
While AI agents are currently a major industry focus, some question the longevity of the trend. However, the view from AWS is that the tools being developed for AgentCore are built on a sustainable pattern. This pattern combines the reasoning capabilities of AI models with the ability to perform real-world actions through tools, a combination expected to endure even as specific technologies evolve.