AWS unveils enhanced tools for AI Agent development

Amazon Web Services (AWS) is enhancing its Amazon Bedrock AgentCore platform, introducing new capabilities to streamline the creation and oversight of AI agents for businesses.

AWS unveiled several new features for AgentCore this Tuesday at its annual re:Invent conference. The announcement highlighted tools for defining agent interaction limits, implementing memory functions, and evaluating agent performance.

A key advancement is the Policy feature within AgentCore. This enables users to define interaction boundaries for agents using natural language. These rules integrate with the AgentCore Gateway, which links AI agents to external tools, to automatically screen each agent's actions and block those that breach the established controls.

According to David Richardson, Vice President of AgentCore, Policy lets developers restrict access to specific internal data or third-party applications such as Salesforce or Slack. Boundaries can also be set for automated tasks—for instance, permitting an AI agent to issue refunds up to $100 autonomously while requiring human approval for larger amounts.

The company also introduced AgentCore Evaluations, a suite of 13 pre-configured systems for assessing AI agents on metrics like correctness, safety, and proper tool usage. This provides developers a foundational toolkit to build upon for their own custom evaluations.

“This directly addresses a major concern people have about deploying agents,” Richardson said regarding the new evaluation tools. “It’s a highly sought-after feature that is often tedious to build from scratch.”

Additionally, AWS announced AgentCore Memory, a new capability that allows agents to maintain a persistent log of user information—such as travel times or hotel preferences—and leverage this data to inform future interactions.

Join the Disrupt 2026 Waitlist

Join the waitlist for Disrupt 2026 to secure your spot when Early Bird tickets are released. Past Disrupt stages have featured leaders from Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla. They were part of over 250 industry experts leading 200+ sessions designed to accelerate your growth and competitive advantage. You'll also have the chance to connect with hundreds of startups driving innovation across all sectors.

Join the Disrupt 2026 Waitlist

Join the waitlist for Disrupt 2026 to secure your spot when Early Bird tickets are released. Past Disrupt stages have featured leaders from Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla. They were part of over 250 industry experts leading 200+ sessions designed to accelerate your growth and competitive advantage. You'll also have the chance to connect with hundreds of startups driving innovation across all sectors.

“Collectively, these three areas represent our ongoing iteration across the different layers of AgentCore,” Richardson stated. “Policy enables interaction with existing systems, Memory makes agents more capable, and the evaluation tools help development teams improve their agents efficiently.”

While AI agents are currently a dominant trend in the industry, with some skeptics questioning their longevity, Richardson believes the foundations being built with AgentCore are durable. He anticipates market trends will evolve, but the platform is designed to adapt.

“Harnessing the reasoning power of these models, combined with their ability to perform real-world actions through tools, represents a sustainable approach,” Richardson explained. “The specific implementation will undoubtedly change over time, and we believe we are well-prepared for that evolution.”

Stay updated with all of TechCrunch's coverage of the annual enterprise technology event here.