Jason Lopatecki is co-founder and chief executive of Arize AI, which builds tools for understanding, evaluating, and improving artificial intelligence systems in production. His work has progressed from analytics-driven advertising infrastructure to AI observability and software agents capable of investigating their own failures.
Lopatecki studied electrical engineering and computer science at the University of California, Berkeley, and co-founded TubeMogul, where he served as chief innovation officer. The video advertising technology company went public in 2014 and was acquired by Adobe in 2016. After the acquisition, he led a machine-learning team at Adobe, confronting an operational blind spot: conventional monitoring could show that software was running while missing whether its models were making sound decisions.
He founded Arize with Aparna Dhinakaran, whom he had met at TubeMogul. His account of Arize’s founding describes the difficulty of understanding machine-learning models after deployment, particularly when their decisions affect consequential applications. As generative AI introduced retrieval errors, unreliable tool use, and multistep agent failures, Arize expanded from model monitoring into tracing, evaluation, and agent reliability. Its products include Phoenix, an open-source observability project launched in 2023, and the commercial Arize AX platform.
How Lopatecki approaches reliable AI
- Turn observability into an improvement loop. Lopatecki treats production traces, logs, evaluations, and repository context as evidence agents can use to investigate failures and prepare fixes. His approach to self-improving agents combines scheduled or event-triggered investigation with human review, particularly when proposed changes exceed straightforward repairs.
- Connect real failures to reviewable engineering work. Within Arize AX, Signal identifies recurring production problems, gathers supporting traces, connects to GitHub repositories, and can prepare issues or proposed pull requests. Lopatecki demonstrated the workflow through Alyx, Arize’s in-product assistant, after a canceled stream and repeated task updates exposed an application failure.
- Evaluate behavior in production. His approach to production agent evaluation layers evaluators onto traces to detect known failure patterns, including prompt injection, while turning newly discovered incidents into future checks. He emphasizes carefully designed observability skills, configurable agent harnesses, sandboxed execution, and customer-controlled enterprise environments.
On August 13, 2026, Dynatrace announced a definitive agreement to acquire Arize for $915 million. The transaction remains subject to regulatory review and other closing conditions; if completed, Lopatecki is expected to continue leading the Arize team.
