Build1 publisher2 min readPublished
Dynatrace pays $915M for Arize to catch AI agents that fail on healthy infrastructure
Dynatrace has closed its $915 million purchase of Arize to trace and evaluate AI agents next to its application monitoring. Its case is that an agent can fail on a healthy stack, so debugging has to follow a bad answer down into the services the agent called.
The Engineer · Build desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Dynatrace first announced its intent to buy Arize in mid-August, after more than two decades as an application performance monitoring vendor.
- Arize came out of stealth in 2020 monitoring machine learning models in production, then expanded into tracing agent behavior and running evaluations.
- Dynatrace's tools have served SRE and platform engineering teams, while Arize's core users are AI engineers and developers.
- Arize's Signal agent reviews production traces, finds recurring problems and proposes likely causes and fixes.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint One product now has to answer two teams with different questions: SREs asking whether services are healthy, and AI engineers asking whether an agent's output was right.
- capability If the traces are joined, a bad agent answer can be followed into the failing API without the separate software-trace lookup Dhinakaran described.
- precedent A $915 million price gives other monitoring vendors a reference point for what an incumbent will pay to own agent tracing and evaluation.
The failure Dynatrace paid for happens on a healthy stack. According to The New Stack, an application can be running perfectly well from an infrastructure perspective while the agent on top of it gives the wrong answer, calls the wrong tool, or fails to finish the task [3]. In that case the infrastructure metrics show a working system [3].
The two halves of the stack meet where the agent calls other software. Aparna Dhinakaran, Arize's co-founder and chief product officer, said an agent typically depends on conventional software components and APIs to carry out a task [14][7]. A fault Arize flags in an agent's reasoning or tool use can therefore start in a service underneath [7]. Describing the split before the deal, she said: "we used to have one side of the coin, the ability to debug all the harness and the LLM-related issues, but if it came to a software issue, we have to then go look at our software traces to figure out the root cause" [8].
Removing that second lookup is the engineering job. In my view the useful product is one trace that runs from the agent's tool call into the service it hit, so a single query reaches the root cause. A shared login and a combined invoice would leave the SRE and the AI engineer reading two systems. The New Stack's interview excerpt does not describe how or when the products will be joined, what the combination will cost, or whether customers have adopted it.
Volume is the second constraint. The New Stack headlined the interview with the line "No human wants to look at billions of traces" [13]. Both companies already point agents at telemetry. Arize has Signal [9]. Dynatrace launched its Davis assistant in 2017 and has since added autonomous SRE agents built to investigate and remediate incidents [10]. An agent that triages agent traces is itself a model, and its proposed causes need the same checking it applies to everything else. "The one thing that we've done on the Dynatrace side for a long time is to invest in the use of AI to power observability," said Steve Tack, Dynatrace's chief product officer [14][11].
Compuware paid $256 million for all of Dynatrace in 2011 [12]. Dynatrace has now paid about 3.6 times that, in nominal dollars, for Arize [1]. The price shows that one incumbent monitoring vendor will spend $915 million to own agent tracing and evaluation [1]. So far, the argument that this belongs in every enterprise observability stack comes from the two product chiefs who explained the deal [14].
What to watch
- Whether Dynatrace ships Arize agent traces and its own service traces as one linked trace, or as two products under one brand.
- Pricing and packaging for Arize after the close, including whether AI engineering teams can still buy it on its own.
- Any published accuracy figures for Signal's proposed root causes on customer production traces.