AWS unveils enhanced features for its AI agent development platform

AWS Expands AgentCore to Enhance Enterprise AI Agent Development

Amazon Web Services is strengthening its AI agent platform, AgentCore, aiming to simplify the creation and oversight of AI agents for business users.

During its annual AWS re:Invent conference, AWS introduced a range of new AgentCore capabilities. These updates include advanced tools for defining agent permissions, enhancing memory functions, and improving agent assessment processes.

Introducing Policy Controls in AgentCore

One significant addition is the Policy feature, which enables organizations to establish clear guidelines for agent behavior using everyday language. Integrated with the AgentCore Gateway—which links AI agents to external services—this function automatically reviews agent actions and blocks any that breach established policies.

With Policy, developers can restrict access to sensitive company data or external platforms like Salesforce and Slack. For example, an agent might be authorized to process refunds up to $100 automatically, but require human approval for larger amounts, as explained by David Richardson, AgentCore’s vice president.

New Evaluation Tools for AI Agents

Another major update is AgentCore Evaluations, a set of 13 ready-made evaluation modules that track aspects such as accuracy, safety, and correct tool usage. These tools give developers a foundation for building custom evaluation systems tailored to their needs.

“This will help address the main concerns people have about deploying agents,” Richardson noted, highlighting that while such evaluation systems are highly desired, they are often complex to develop from scratch.

AgentCore Memory: Smarter, Context-Aware Agents

AWS is also adding AgentCore Memory, a feature that allows AI agents to accumulate and reference user information—such as travel schedules or accommodation preferences—over time, enabling more personalized and informed interactions in the future.

“With these three advancements, we’re continuing to evolve AgentCore at every level,” Richardson said. “Policy enables integration with existing systems, AgentCore Memory empowers agents with context, and the evaluation tools support ongoing development.”

Looking Ahead: The Future of AI Agents

Although AI agents are currently a hot topic in the tech world, some are skeptical about their longevity. Richardson, however, believes that the innovations within AgentCore will remain valuable even as industry trends shift.

“Leveraging the reasoning abilities of these models, combined with their capacity to interact with real-world tools, creates a lasting framework,” Richardson explained. “While the specifics will evolve, we’re confident in our ability to adapt.”

Stay updated with all the latest news from TechCrunch’s coverage of the annual enterprise technology event.