Dynatrace has completed its acquisition of Arize, bringing specialised AI observability capabilities for tracing, evaluation and experimentation into its broader application and infrastructure monitoring portfolio. The deal connects Arize's tools for examining how AI applications and agents behave with Dynatrace's platform used by SRE, platform engineering and operations teams. Dynatrace sees the combination as a foundation for observability across the full AI lifecycle, although deeper product integration is still ahead.
The acquisition was first announced in August, and with the transaction now complete, both companies are beginning work on a joint product and platform roadmap. Dynatrace says future priorities will be shaped by feedback from customers and developers. For now, the focus is on identifying where the two platforms can be connected most effectively.
Bringing AI Behaviour Into Traditional Observability
Traditional observability is very good at showing whether applications, infrastructure and services are healthy, but that alone does not explain whether an AI system is behaving correctly. Steve Tack, Chief Product Officer at Dynatrace, said technical health does not by itself determine whether an AI system is working as intended. Teams also need context across the lifecycle to understand what happened, where a problem started, what it affected and how the findings should influence the next iteration.
This distinction is especially important for AI applications and autonomous agents. A service can remain fully available while still producing an incorrect answer, using outdated context or interacting with the wrong data. In these situations, conventional infrastructure telemetry may show no obvious failure even though the AI outcome itself is wrong.
Arize Focuses On Agent Behaviour And Outcomes
Arize provides workflows that allow AI engineers to inspect agent trajectories in detail. Teams can examine model calls, retrieved context, tool usage, execution performance and cost, giving them a clearer picture of how an AI application arrived at a particular result. They can also run evaluations, compare experiments and use those findings to guide future changes.
This is particularly useful for complex agentic systems where a single user request may trigger several models, data sources and external tools. Tracing that entire path can reveal whether an issue came from the model itself, a retrieval step, a tool call or another part of the workflow. That level of visibility becomes increasingly important as enterprise AI systems become more autonomous.
Dynatrace Adds The Wider Operational Context
Dynatrace brings a different layer of visibility by monitoring the surrounding applications, infrastructure, services, user experiences and business processes. The proposed integration would connect this operational context with the detailed AI evidence collected by Arize. Together, the goal is to help teams understand not only what an AI agent did, but also how that behaviour interacted with the wider production environment.
For example, an AI agent might complete a workflow successfully from an infrastructure perspective while still updating the wrong record or making a decision based on stale information. In such a case, application health metrics alone would not provide enough context. Combining operational telemetry with AI-specific traces and evaluations could make those issues easier to diagnose.
Evaluation Does Not Stop After Deployment
Dynatrace and Arize are also emphasising that AI evaluation should continue after an application reaches production. Models, prompts, retrieval methods, data sources, tools and agent workflows can all change over time, which means behaviour that was acceptable during development may shift later. Continuous evaluation becomes especially important when AI systems are updated frequently.
Findings from production investigations can then feed back into development. Teams can turn those findings into datasets, regression tests, evaluators or new experiments for future iterations. Connecting this feedback loop with real-world production context is one of the main areas Dynatrace plans to explore, although this remains part of the future product roadmap rather than a completed capability.
Closing The Loop Between Development And Production
Stephen Elliot, IDC Group Vice President for Software Development and IT Operations, said observability is becoming increasingly important as enterprises deploy more AI agents. He noted that bringing evaluation and observability together can help close the gap between building AI applications and operating them reliably in production. This could allow teams to identify problems earlier and resolve them more quickly.
The idea is to create a more continuous engineering loop between AI development and production operations. Instead of treating model evaluation and infrastructure monitoring as separate activities, teams could potentially analyse both within a more connected workflow. That becomes more valuable as AI systems grow in complexity and become more deeply embedded in business processes.
Phoenix And OpenInference Remain Open Source
Arize's open-source Phoenix project will continue to play an important role after the acquisition. Phoenix supports tracing, evaluation, failure investigation, dataset curation and experiment comparison, while Arize AX provides similar workflows through a managed platform that spans development and production environments. Both products remain available following the completion of the deal.
Phoenix is built on OpenTelemetry and OpenInference. Arize originally created OpenInference as an open-source instrumentation project and set of semantic conventions designed specifically for AI observability. In June 2026, OpenTelemetry formally accepted a code grant covering OpenInference GenAI instrumentation, with that work now being integrated progressively into OpenTelemetry's own GenAI instrumentation project.
Dynatrace Plans To Preserve The Open-Source Approach
OpenInference will continue as an open and OpenTelemetry-compatible project, while Dynatrace says it intends to support Arize's existing open-source approach. The company is now evaluating how Phoenix and Arize AX workflows could be connected with the operational context already available through the Dynatrace platform. Tack said near-term product work will focus on identifying where those integrations can deliver the most value across the AI lifecycle.
For users already working with Arize products, there is no immediate requirement to migrate. Arize AX and the Dynatrace platform continue to be offered as standalone products, while Phoenix remains available as an open-source project. This gives existing users continuity while the combined roadmap is developed.
Arize Leadership And Team Join Dynatrace
Arize CEO Jason Lopatecki, Chief Product Officer Aparna Dhinakaran and the wider Arize team are joining Dynatrace as part of the acquisition. Their expertise in AI observability and evaluation will now become part of Dynatrace's broader product organisation. This should give Dynatrace a stronger foundation as it expands beyond traditional application and infrastructure monitoring into more AI-specific operational challenges.
The acquisition also reflects how observability itself is evolving. As AI agents become part of production systems, teams need to understand not just whether services are available, but whether those agents are making correct decisions and using the right context. That requires a different level of visibility from conventional infrastructure monitoring alone.
Final Thoughts
Dynatrace's acquisition of Arize is a significant move towards combining traditional observability with the specialised tracing and evaluation required by modern AI systems. Arize brings visibility into models, agent behaviour, retrieved context and tool usage, while Dynatrace contributes a broader view across applications, infrastructure, users and business processes. Together, the two platforms could provide a more complete way to investigate AI failures that might otherwise remain invisible in standard monitoring tools.
The integration itself is still a work in progress, so the most important changes will come as Dynatrace and Arize develop their joint roadmap. For now, Arize AX, Dynatrace and Phoenix continue to operate independently, while the open-source work around Phoenix and OpenInference remains intact. The long-term goal is clear: create an observability platform that follows AI systems from development and experimentation all the way through production operation and continuous improvement.
Comments 0