Dynatrace buys Arize because service stopped meaning the right answer

Dynatrace completed the acquisition of Arize, bringing AI observability tools for tracing, evaluation and experimentation. The announcement came out on October 2. The intent to buy had already been disclosed in August.

The deal brings together two engineering worlds. However, on one side, tools to examine applications and AI agents. On the other, the observability platform used by SRE and platform engineering teams.

The integration of the products, however, is still underway.

The thesis behind the deal

Steve Tack, chief product officer at the acquiring company, summed up the problem. According to him, technical health alone no longer determines whether an AI system works as expected.

Notice the detail that separates the two worlds. Moreover, an application can return an answer that is plausible and incorrect without generating any error.

The same goes for agents. However, one of them can complete a flow using outdated context or updating the wrong record, while all supporting services appear healthy.

Therefore, latency and error rate metrics tell only part of the story.

Dynatrace gains visibility into the agent's trajectory

The acquired company's flows allow inspecting the complete trajectory. This includes model calls, retrieved context, tool use, performance and cost.

Moreover, teams run evaluations and compare experiments. However, the results guide subsequent changes to the application or the agent.

The acquiring platform, in turn, brings the surrounding context. It covers applications, services, infrastructure, user experience and business processes.

The proposed integration links these operational records to the evidence used in evaluating behavior.

Tack argues that application and infrastructure telemetry alone fails to explain an agent's actions. The cost of the trajectory is also left out of that accounting.

Why evaluation continues after deployment

Here is the most useful point for anyone operating AI systems. Changes happen all the time in models, prompts, retrieval strategies, tools, data and agent flows.

Consequently, behavior changes without any change to the application's code.

Moreover, the proposed path closes the loop. Investigation findings become datasets, regression tests, evaluators or experiments for the next iteration.

Stephen Elliot, vice president at IDC, reinforced this reading. For him, uniting evaluation and observability closes the loop between building AI applications and running them reliably in production.

The open side remains standing

Phoenix remains an open-source project. However, it covers tracing, evaluation, failure investigation, dataset curation and experiment comparison.

The managed version is called Arize AX. It delivers these same flows covering development and production.

The technical foundation matters to those who avoid vendor lock-in. Phoenix runs on OpenTelemetry and OpenInference.

This second project started as open instrumentation and a set of semantic conventions for AI observability.

In June 2026, OpenTelemetry formally accepted a code donation for GenAI instrumentation coming from it. This code has been gradually incorporated into the GenAI instrumentation project.

OpenInference remains open and compatible with the standard.

What this changes in your stack

First, treat evaluation as part of operations. Running an evaluator only before deployment lets regressions caused by a model swap slip through.

Second, instrument the entire trajectory. Model call, retrieved document and invoked tool need to appear in the same trace.

Third, log cost per run. However, this data supports architecture decisions and vendor negotiation.

Moreover, prefer open conventions. OpenTelemetry-based instrumentation allows swapping the visualization platform later.

Finally, turn incidents into tests. A real investigated case becomes the best regression set you can have.

What to watch going forward

Jason Lopatecki, CEO of the acquired company, Aparna Dhinakaran, CPO, and the whole team remain with the acquiring company. Both platforms remain available independently.

Meanwhile, the companies are starting to build a joint roadmap. Moreover, priorities, according to the announcement, come from customers and developers.

Follow our profile on Instagram!