Research Finder
Find by Keyword
Dynatrace and AgentCore: Is Observability the AI Agent Bottleneck?
End-to-end tracing, cost analysis, and compliance monitoring for autonomous AI workflows using OpenTelemetry and AWS Bedrock AgentCore are key themes.
19/11/2025
Key Highlights:
- Dynatrace announced the general availability of its deep integration with Amazon Bedrock AgentCore.
- The integration aims to deliver full, end-to-end observability for complex, multi-step AI agentic workflows.
- It provides unified tracing, real-time cost analytics, and guardrail monitoring to ensure compliance and reliability.
- This support is engineered to help organizations confidently scale autonomous AI applications on AWS.
- The solution utilizes OpenTelemetry signals enriched with specialized Generative AI semantic attributes.
The News
Dynatrace announced that its integration with Amazon Bedrock AgentCore is now generally available for AWS customers. This collaboration delivers real-time visibility into the behavior and performance of autonomous agents and their interactions across AWS services. It provides essential capabilities for monitoring, debugging, optimizing, and auditing complex agentic workflows. The move reflects an understanding that production-grade AI agents require enterprise-grade operational transparency. Find out more in the press release.
Analyst Take
The conversation around enterprise generative AI has shifted focus. For the past year, the industry fixated on models, token counts, and retrieval-augmented generation (RAG) patterns. Now, the complexity of moving from impressive proofs-of-concept to scaled, autonomous systems is taking center stage. Agents are the next evolution of this technology, yet they introduce profound operational challenges far beyond traditional application monitoring. This is where Dynatrace is placing its strategic bet, positioning observability not as a passive metric collector, but as the foundational layer of trust for agentic architectures.
Enterprise teams are discovering that agentic failure modes look nothing like conventional software failures. Agents make decisions that are influenced by context, memory, and prior tool interactions, which means the root cause is often conceptual rather than technical. That shift places a premium on systems that can correlate reasoning traces with infrastructure telemetry to reconstruct the agent’s decision process in real-time.
Amazon Bedrock AgentCore is the platform designed to streamline this complex transition. Agents are inherently non-deterministic, executing multiple steps, invoking various tools, and communicating with other services in a non-linear fashion. These emergent behaviors create massive blind spots for traditional monitoring tools. If a business-critical agent fails or goes off-script—perhaps exceeding budget or violating a safety guardrail—identifying the root cause involves tracing through model reasoning, tool invocations, and multi-agent communication chains. This is a formidable technical challenge.
This is why many early agent pilots stall during the move to production. The operational surface area expands dramatically with latency, cost drift, policy violations, hallucinated tool calls, and unexpected decision loops all surfacing once agents encounter real-world variability. Without deep observability, these issues look like random failures, but with trace-based insight they resolve into predictable, diagnosable patterns that can be governed and optimized.
Dynatrace’s immediate GA support for AgentCore is a smart and timely move that solidifies its partnership with AWS and differentiates its offering in the highly competitive observability landscape. The company is stepping into a greenfield segment: enterprise agent operations. While competitors focus on basic LLM-layer telemetry, Dynatrace targets the complete agent lifecycle, from orchestration and tool execution to compliance and cost management. This approach understands that the value of an agent is tied directly to its reliability and auditability in a production environment. Customers like Cox Automotive, who are already using AgentCore, need solutions that can confidently manage systems, transforming core business processes.
What Was Announced
The new capabilities are architected to embed observability directly into the agent’s execution path. The integration is built on the OpenTelemetry standard, specifically leveraging GenAI semantic attributes to enrich the collected data. This means that every step in the agent’s decision-making process—the initial user prompt, the model’s reasoning trace, and all subsequent tool calls—is captured as a distributed trace.
The Dynatrace AI Observability app is designed to convert this agent telemetry into actionable insights. Teams gain visibility into agent reliability and responsiveness, right down to the trace level. The platform aims to deliver intelligent alerting capabilities based on key operational metrics, such as latency spikes or deviations in token consumption. A crucial feature is the ability to visualize agent-to-service interactions through a real-time topology map. This map allows developers to immediately understand the dependencies and data flows across the entire stack, which is critical for debugging complex, multi-step agent workflows.
Furthermore, the solution includes mechanisms to oversee compliance and governance in AI-driven systems. By tracking specific guardrail metrics, such as toxicity scores, filtered content, and denied topics, teams can ensure that agents operate within organizational policies and emerging regulatory guidelines, such as those anticipated from the EU’s Artificial Intelligence Act. Automated cost calculations for each agent request are also designed to help monitor and manage consumption, allowing teams to set up error budgets for performance and financial control. The architecture provides deep visibility, enabling teams to pinpoint issues not just to a service, but to an exact prompt, token count, or tool integration step.
Our analysis suggests this is not merely an integration but an extension of Dynatrace’s core Davis AI capabilities into the realm of autonomous systems. It is an acknowledgement that AI must observe AI. By being one of the first to provide this level of comprehensive, trace-based visibility for a major cloud provider's agent platform, Dynatrace is aiming to become the default operational control plane for agentic AI on AWS. The complexity of these new applications demands robust tooling.
Looking Ahead
The market is quickly delineating between basic LLM monitoring and comprehensive agent observability. The Dynatrace announcement underscores that the ability to scale autonomous systems is fundamentally constrained by the ability to monitor and debug them. It is not enough to know an agent failed; you must know why it decided to fail and where in the chain the logic broke down. This focus on causal AI analytics applied to non-deterministic agent behavior is the high-leverage move.
The broader implication is that observability is becoming inseparable from AI safety and financial governance. As enterprises adopt autonomous workflows, the operational layer becomes the enforcement point for policies, cost constraints, and risk posture. This expands observability’s role from monitoring system health to actively shaping how AI systems behave at runtime.
This move positions Dynatrace directly against the tectonic shift toward integrated cloud operations tooling. We see a similar trend in their work with Microsoft Azure SRE Agent, indicating a clear strategy to embed their observability fabric into the control planes of the hyperscalers as they roll out their own agentic and autonomous solutions. This multi-cloud strategy is prudent.
Our perspective is that competitors who rely solely on log or metric scraping will struggle to offer the granular, trace-level data necessary for debugging prompt chains and tool calls. Agentic AI necessitates a depth of context that only end-to-end tracing can provide. The key trend that we are going to be looking out for is how quickly other observability vendors can build out equivalent support for AgentCore and other platforms like Azure’s. HyperFRAME will be tracking how the company performs on customer adoption and feature parity in future quarters, specifically around the robustness of their automated cost and compliance reporting for live, autonomous workflows. The real differentiator will be the speed and accuracy with which the observability platform can flag and auto-remediate agent behavior that risks financial or regulatory exposure. This is the operational intelligence customers need.
Stephanie Walter | Practice Leader - AI Stack
Stephanie Walter is a results-driven technology executive and analyst in residence with over 20 years leading innovation in Cloud, SaaS, Middleware, Data, and AI. She has guided product life cycles from concept to go-to-market in both senior roles at IBM and fractional executive capacities, blending engineering expertise with business strategy and market insights. From software engineering and architecture to executive product management, Stephanie has driven large-scale transformations, developed technical talent, and solved complex challenges across startup, growth-stage, and enterprise environments.
Steven Dickens | CEO HyperFRAME Research
Regarded as a luminary at the intersection of technology and business transformation, Steven Dickens is the CEO and Principal Analyst at HyperFRAME Research.
Ranked consistently among the Top 10 Analysts by AR Insights and a contributor to Forbes, Steven's expert perspectives are sought after by tier one media outlets such as The Wall Street Journal and CNBC, and he is a regular on TV networks including the Schwab Network and Bloomberg.