Amazon Bolsters AgentCore with Governance, Evaluation, and Memory Enhancements
Event summary
- Amazon announced new features for Amazon Bedrock AgentCore at AWS re:Invent on December 2, 2025.
- AgentCore now includes Policy controls to block unauthorized actions, operating outside of agent code.
- AgentCore Evaluations provides developers with tools to continuously assess agent quality and behavior.
- AgentCore Memory introduces episodic functionality, enabling agents to learn from past experiences.
- A diverse set of organizations, including Archera, Cohere Health, and S&P Global Market Intelligence, are utilizing AgentCore.
The big picture
Amazon’s enhancements to AgentCore represent a strategic move to address growing concerns around AI safety and control. By embedding governance and evaluation tools directly into its agentic AI platform, Amazon is positioning itself as a leader in responsible AI development. This focus on control and observability is likely to become increasingly important as generative AI models are deployed in regulated industries and mission-critical applications.
What we're watching
- Governance Dynamics
- The adoption rate of AgentCore’s Policy controls will indicate the industry’s prioritization of AI governance and risk mitigation, potentially impacting the broader generative AI landscape.
- Evaluation Adoption
- The extent to which developers leverage AgentCore Evaluations will reveal the industry’s commitment to ongoing AI quality assurance and refinement, which could influence the development lifecycle of agentic AI applications.
- Memory Scalability
- The ability of AgentCore Memory to handle increasingly complex and voluminous episodic data will be crucial for its long-term utility and adoption across diverse use cases.
Related topics