Amazon обновила Bedrock AgentCore: агенты получили доступ к корпоративным знаниям и мониторингу

Amazon Web Services expanded Bedrock AgentCore — an enterprise platform for creating, integrating, and optimizing AI-agents. The update closes three key gaps that teams encounter when transitioning from prototype to production.

Knowledge from Any Source

One of the main challenges of production-agents is limited access to up-to-date data. Bedrock AgentCore now allows agents to connect directly to multiple types of sources:

• Corporate knowledge bases (internal documents, CRM, wiki)
• Open web — for searching current information in real time
• Paid and licensed data sources
• Organization's structured repositories (S3, databases)

This allows the agent to answer not only based on model training, but also rely on fresh and business-specific data — without additional fine-tuning.

Debugging in Production

When an agent makes a mistake, finding the cause can be difficult: multi-step reasoning chains are opaque. AgentCore adds tools that help teams track agent behavior in real time and find failure points. Teams can see which step in the chain produced an incorrect result, which knowledge source was involved, and where the agent lost context. Essentially, this is an observability stack for agent systems — an analogue of tracing in microservices, but applied to AI logic.

Control That Scales with the Agent

The more capabilities an agent has, the harder it is to control. Amazon solves this problem through built-in governance mechanisms that scale alongside the tasks.

"We help teams build more powerful agents faster, manage them through scalable control mechanisms, and continuously improve them,"

AWS's official announcement states.

Security policies, tool restrictions, and data access rules are set centrally and applied automatically — regardless of how complex the agent chain becomes.

What This Means

Bedrock AgentCore is transforming from an experimentation tool into a full-fledged production stack for enterprise agents. For teams already building agent systems on AWS, this lowers the barrier to entry for enterprise deployments and eliminates the need to assemble monitoring and control from separate components manually.