While modern artificial intelligence models possess remarkable reasoning capabilities, multi-step workflow planning, and nuanced response generation, many enterprise agents operate far below their true potential. This performance gap is rarely an issue of core intelligence; rather, it stems from a lack of continuous feedback and contextual access.
For instance, customer support agents frequently fail to resolve inquiries if critical refund policies are locked within siloed SharePoint folders. Similarly, research agents often deliver incomplete market briefs due to data limitations that extend beyond their training parameters, while financial advisor agents may offer suboptimal recommendations when isolated from real-time market data. Furthermore, most organizations still lack systematic methodologies to evaluate whether their deployed agents are improving or deteriorating over time.
To address these production hurdles, AWS has introduced major feature updates to Amazon Bedrock AgentCore, a comprehensive platform designed for building, connecting, and optimizing AI agents. The latest enhancements close critical operational gaps by establishing native connections to organizational, web-aligned, and premium paid knowledge. Additionally, the platform provides advanced diagnostic tools to resolve production anomalies while maintaining deterministic security controls that scale alongside agent capabilities.
Tri-Layer Knowledge Integration: Organizational, Web, and Paid Data
Amazon Bedrock AgentCore now grants agents native access to three distinct layers of knowledge, significantly broadening their operational reach:
- Organizational Knowledge Layer (Bedrock Managed Knowledge Base of Amazon): Previously, the task of linking agents to organizational knowledge dispersed in SharePoint, Confluence, Google Drive, and other wiki systems used for internal documentation took months of data engineering efforts. The use of Bedrock Managed Knowledge Base solves this problem through the automation of pipeline handling. Users can just connect their unstructured datasets, after which the agent core takes care of vector stores, embeddings, re-ranking models, and the scaling issues involved. With the help of the advanced agentic retriever, the solution generates queries and finds relevant concepts in various knowledge bases.
- World Knowledge Layer (Web Search on AgentCore): To keep pace with shifting market regulations and competitive landscapes, developers can leverage the new Web Search tool. Built on the secure Amazon infrastructure powering Alexa+, Amazon Quick Suite, and Kiro, this tool queries the live web while keeping sensitive data entirely within the client’s protected AWS environment. It utilizes a multi-source grounding methodology that fuses public web data with Amazon’s proprietary knowledge graph, incorporating real-time metrics like stock market fluctuations and verified facts.
Commenting on the integration, Masahiro Oba, Senior General Manager at Sony Group Corporation, stated: “At Sony, we’re building an enterprise AI agent platform on AgentCore where teams across business units can develop, share, and reuse AI agents – from knowledge assistants to workflow automation agents – each tailored to their needs. Our enterprise knowledge is distributed across repositories such as SharePoint, Confluence, and Amazon S3, and includes complex documents such as PDFs, presentations, and spreadsheets with charts and tables. Now that Bedrock Managed Knowledge Base and Web Search are available in AgentCore, we can equip agents with advanced retrieval and live web grounding with a consistent governance model, without building these capabilities from scratch. This accelerates our vision of transforming how people work, with AI as a catalyst, at scale.”
- AgentCore Payments Layer (AgentCore Payments & AWS WAF AI Traffic Monetization): High-end data sets, APIs, and financial feeds are not commonly accessible at no cost. In order to make use of paid knowledge, AgentCore payments (which is currently in preview mode) provides agents the ability to discover, acquire, and transact premium content within their execution cycles. As a provider, AWS WAF AI traffic monetization helps to control agent behavior by allowing content owners to either permit, deny, or monetize agent interactions through a secure path.
Also Read: Dataiku Expands Enterprise AI Capabilities With Cobuild on Snowflake
Data-Driven Optimization and Continuous Improvement
Identifying operational failures in probabilistic AI systems is notoriously difficult. The most disruptive bugs often bypass traditional engineering dashboards; an agent might falsely confirm an unexecuted order modification or hallucinate inventory availability during an API timeout while telemetry logs display a perfect success rate.
To transition from guesswork to data-backed enhancement, AgentCore has launched new optimization workflows:
- Advanced Diagnostics (Preview): The platform delivers deep insights into user intent, agent failures, and session trajectories across hundreds of parallel workflows. These automated failure summaries expose hidden behavioral anomalies, clarify their root causes, and prioritize them by user impact.
- Validation and A/B Testing (General Availability): Developers can leverage automated recommendations to optimize system prompts and tool descriptions. These updates can be benchmarked via batch evaluations against custom test datasets. Furthermore, live production traffic can be partitioned for controlled A/B testing, confirming system reliability before final deployment.
Highlighting the practical value of these capabilities, Kazumi Matsuda, Senior Manager, AI Promotion Department at FUJISOFT, noted: “At FUJISOFT, we’re building AI agents to accelerate software development and operations. Our framework, Character Capsule, packages agent roles, skills, and procedures as reusable capsules that run on local coding tools like Copilot and Kiro, or scale to multi-agent orchestration on AgentCore. As we deployed more agents, our biggest challenge was the silent failures that looked fine but surfaced later, and fixing them was guesswork. The optimization capabilities in AgentCore changed this. They analyze our production traces to surface failure patterns, explain why they happen, and rank them by impact. We then get recommendations to improve our prompts and tool descriptions, and A/B test them on live traffic before committing. Agent improvement is now a continuous loop grounded in data, not trial and error.”
Deterministic Governance and Scalable Security
As a result of their probabilistic nature, AI models are prone to security weaknesses, including prompt injection and memory poisoning. In an effort to address this problem, AWS has embedded Amazon Bedrock Guardrails in AgentCore’s gateway. Using this deterministic framework, the system checks each of the agent’s actions for any malicious activity, harmful content, and leakage of sensitive information irrespective of its current context to stop the agent from crossing security boundaries. Further updates will enable the integration of third-party detections from security vendors such as Check Point, Zscaler, Rubrik, Netskope, and SentinelOne.
Accelerating Development via the AgentCore Harness
Building a resilient framework to handle orchestration loops, state persistence, and error recovery usually consumes the bulk of engineering roadmaps. The now generally available AgentCore harness addresses this by providing a fully managed runtime environment. Developers can configure models, tools, and execution parameters via a centralized file, allowing AgentCore to provision an operational agent complete with a managed filesystem, multi-session memory, and integrated web-browsing capabilities within minutes.
Importantly, this managed harness decouples the orchestration logic from specific underlying models. Enterprises remain free to swap foundational models mid-session without rebuilding core application logic.
Reflecting on the infrastructure synergy, Omar Paul, VP of Product at Twilio, remarked: “Twilio’s customers are building AI agents that work across voice, messaging, and digital channels, with real-time intelligence and persistent memory that make every interaction feel like a conversation. By combining AgentCore harness with Twilio Conversations, developers can go from idea to live agent without rewiring infrastructure. The best customer experiences happen when great AI and great communications infrastructure are built together.”
