According to a report by SiliconANGLE, enterprise observability is entering a new phase as artificial intelligence transforms both the applications companies need to monitor and the way operations teams respond when incidents occur. Traditional observability platforms were built around deterministic software and telemetry, such as logs, metrics, and traces. In contrast, AI applications and agents behave differently, producing outputs that can vary even when provided with similar inputs. At the same time, enterprises expect observability platforms to move beyond merely identifying problems and provide sufficient context to enable humans and AI agents to diagnose, remediate, and potentially act directly on those issues.
This transition provides the backdrop for Dynatrace’s acquisition of Arize AI, a move that integrates AI observability, evaluation, and agent monitoring capabilities into Dynatrace’s broader application observability platform. In an episode of theCUBE Research’s AppDevANGLE podcast, Practice Lead and Principal Analyst Paul Nashawaty spoke with Steve Tack, chief product officer of Dynatrace, and Aparna Dhinakaran, co-founder and chief product officer of Arize AI, about the convergence of application observability and AI observability and what it means for enterprise operations. Tack noted that the world has shifted significantly and that AI brings new problems and new domains to the space.
From deterministic software to nondeterministic systems
One of the central shifts is taking place inside the applications themselves. Traditional software generally produces predictable outcomes, whereas AI-powered applications—particularly those incorporating large language models (LLMs) and autonomous agents—introduce nondeterministic behavior that makes troubleshooting more difficult. This means that observability can no longer focus solely on whether an application is available or whether infrastructure is operating within expected performance thresholds. Teams must also understand whether the AI system produced the intended response and whether the quality of that response met expectations.
Dhinakaran explained that the evaluation phase is no longer just about whether a result is right or wrong, but about actually measuring the quality of responses, which represents a fundamentally different problem. Arize built its platform around this challenge, providing tools for tracing, evaluating, and improving AI applications and agents. Its open-source platform, Phoenix, is used by more than 4,000 enterprises according to Dhinakaran, while the Arize AX product provides a managed environment designed for teams operating AI systems at production scale. For Dynatrace, these capabilities extend observability into an application layer that is growing in importance as enterprises transition AI projects from experimentation into production.
Sharing context between AI telemetry and application telemetry
AI applications rarely operate independently. Agents call application programming interfaces (APIs), interact with databases, depend on cloud infrastructure, and connect to broader enterprise systems. This creates a troubleshooting challenge when the AI behavior being investigated represents only one part of a much larger software stack. Dhinakaran noted that Arize customers increasingly requested stronger connections between AI telemetry and traditional application and production telemetry, while Dynatrace customers were asking for deeper AI observability and evaluation capabilities.
Unifying the two environments could provide developers, site reliability engineers (SREs), platform teams, AI engineers, and data scientists with a shared view of what is happening across the application stack. Dhinakaran emphasized that agent systems and software systems are joined at the hip, and having the ability to debug agents alongside access to the complete context of the software they use to call tools or the underlying infrastructure behind them enables the development of better products.
This shared context could also help address another persistent challenge in observability: tool sprawl. According to research cited by Nashawaty during the conversation, 75% of organizations use between six and 15 tools for observability. When enterprises add monitoring, evaluation, and governance systems for AI, there is a risk that AI will create an additional isolated operational layer rather than reducing complexity. Tack argued that combining application observability and AI observability provides organizations with a more complete system-level view, rather than forcing teams to piece together information across disconnected platforms.
From dashboards to autonomous operational action
The broader shift may focus less on what observability platforms monitor and more on who—or what—consumes the information. For years, observability was primarily designed around engineers examining dashboards, responding to alerts, and manually troubleshooting incidents. AI agents create the possibility of a different operational model, in which telemetry becomes context that software agents can consume directly. Dhinakaran noted that observability is no longer about humans looking at dashboards, metrics, and logs, but about action.
This perspective alters the value of observability data: instead of simply explaining what happened, telemetry can become part of the reasoning layer that agents use to identify problems, recommend changes, or initiate remediation. The transition also raises the bar regarding precision and context, as autonomous operations only work if organizations trust the information feeding those decisions. Tack pointed out that the ability to provide precise analytics and trustworthy answers will be essential as enterprises give agents greater responsibility. Dynatrace is advancing in this direction through its broader AI and automation strategy, which includes Dynatrace Intelligence and its BlueBox AI offering for agentic development and SRE workflows, with Arize adding deeper evaluation and observability capabilities around the AI systems participating in those processes.
Changes in how software development teams operate
The acquisition also reflects a broader shift in software development itself, where AI agents are increasingly used not only inside applications but also to build, test, troubleshoot, and operate those applications. Tack described a future in which architects may spend less time working directly inside development environments and more time coordinating groups of specialized agents, turning observability into part of the feedback loop between autonomous development and production operations. For enterprises, this could bring observability closer to an operational intelligence layer spanning application development, AI evaluation, infrastructure, and automated remediation. The challenge will be ensuring that automation progresses alongside the governance, reliability, and confidence required before handing meaningful operational decisions over to agents.