“No human wants to look at billions of traces”: Dynatrace bought Arize because agents need a new kind of observability

Dynatrace has spent years watching software from the bottom up: apps, services, infrastructure, and the security layers around them. Arize came at the problem from the top, with tools for tracing, evaluating, and debugging models and AI agents. Now the two are under one roof, and that matters because agents can fail even when the underlying system looks fine.

That gap is the whole point of the deal. Dynatrace announced its intent to buy Arize in mid-August for $915 million, and the acquisition has now closed. Steve Tack, Dynatrace’s chief product officer, says the company’s long habit of using AI inside observability made Arize feel like a natural fit. Dynatrace launched Davis in 2017, then layered in causal AI, predictive and generative AI, and more recently autonomous SRE agents. Arize, meanwhile, built its name after emerging from stealth in 2020 as an AI observability startup focused on production model monitoring.

The overlap is obvious once you look at how modern agents work. They lean on regular software components and APIs to do anything useful, which means a failure can sit in the model, the tool use, or the old-fashioned service underneath it. Aparna Dhinakaran, Arize’s co-founder and chief product officer, says that used to force teams to jump back and forth between LLM traces and software traces. With both platforms together, she says, the idea is to debug across both layers instead of treating them as separate problems.

That’s also why this deal is about more than prettier dashboards. Dynatrace and Arize both think observability is moving from passive inspection toward action. Dhinakaran says no human wants to inspect billions of traces, and that agents will increasingly read telemetry, spot recurring failures, and even open pull requests. Arize already does a version of that with Alyx, its own assistant, where Signal reviews traces and opens fixes that Arize accepts roughly 65% to 70% of the time.

Dynatrace says Arize’s Phoenix and OpenInference were part of the attraction, and both companies say Arize will continue to exist as a standalone product for now. But the direction is pretty clear: build AI applications, watch them in production, then let software help repair itself. Humans can keep the final review. For now, anyway.