Dynatrace’s $915M Arize deal bets AI agents are just another app to monitor
The New Stack Matthew Burns
Dynatrace spent $915M to buy Arize and fold AI-agent tracing into its app monitoring. The bet: agents are just software, so they need the same blame-and-cost tools as everything else.
Based on reporting by The New Stack, Matthew Burns — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Dynatrace has closed its $915 million purchase of Arize, tying AI-agent tracing and evaluation into the company’s existing view of applications and the infrastructure underneath them. The move also pulls an independent AI observability vendor off the market, including one that Datadog had backed.
Sean O’Dell, a principal product marketing manager at Dynatrace, framed the deal around a simple idea: an AI app is still an app. If it goes sideways, someone has to know what the outage cost and what actually broke, whether that’s a bad response, a hallucination, or something deeper in the stack.
That matters because these systems do not live alone. O’Dell pointed out that AI apps still sit inside a wider mess of legacy software and, in many cases, mainframes. Until now, that often meant switching between two tools: one for the AI layer and another for the underlying software problem that caused the issue in the first place.
Arize brings more than just tracing. It also brings Phoenix, its source-available tool for tracing and testing AI apps, plus the developers who already use it. O’Dell called that community hard to find because the space is still so early.
Both companies have agents that can act on the data they see. Dynatrace’s Bluebox reads code in GitHub, GitLab or Bitbucket, compares it with production data and suggests a fix. Arize’s Signal does something similar for Alyx, Arize’s AI assistant, and Arize says it accepts about 65% to 70% of the pull requests Signal writes.
But the automation line still has a human on it. O’Dell said teams can move toward being “a human out of the loop” with Bluebox, yet he still tells customers to validate any fix before approving it. That restraint sounds old-fashioned, but it also sounds familiar to anyone who has watched ops teams trust software with production and then spend the rest of the week cleaning up the mess.
My take — AI-written commentary, not fact-checked reporting
The loudest thing here is not the price tag. It’s the very non-sexy belief that AI agents are just another production system with failure modes, costs, and blame to assign. That’s the right bet, and it’s a nice reminder that the future of AI is still partly held together by old ops habits and a healthy distrust of shiny bots.
Read more about this at: The New Stack