Arize AI hopes it has first-mover advantage in AI observability

In the ever-evolving world of cloud software, platforms like Dynatrace and ServiceNow have long been instrumental in spotting and resolving code errors or system failures. Now, Arize AI is applying a similar strategy to the realm of artificial intelligence, offering an observability platform specifically designed for AI models and applications.

Arize's platform assists companies in evaluating their AI products during development and continuously monitoring them for errors and issues once they're operational. It supports a wide range of AI applications, from machine learning and computer vision to the burgeoning field of generative AI.

Jason Lopatecki, Arize's co-founder and CEO, shared with TechCrunch that the company employs a "council of judges" method to monitor and evaluate AI. This involves using multiple AI models to assess each other, which Lopatecki humorously described as "very meta," alongside human oversight.

The concept for Arize originated from Lopatecki's experience at TubeMogul, a brand advertising company acquired by Adobe for over $500 million in 2016. At TubeMogul, AI was central to operations, and any malfunction was a significant issue due to the technology's complexity. Aparna Dhinakaran, Arize's co-founder and CPO, who connected with Lopatecki through TubeMogul, faced similar challenges while developing language models without adequate tools for testing and evaluation.

Both founders recognized the critical role AI would play across various organizations and the inherent difficulties in understanding and troubleshooting it. This realization led them to launch Arize in 2020, initially focusing on predictive machine learning. From a mere idea at its inception, Arize has grown significantly over the past five years, now supporting a broad spectrum of AI technologies, from AI agents to generative AI.

Lopatecki described the last two years as a period of explosive growth for Arize, attributing this to the increased accessibility of AI. "Everyone's a prompt engineer. Every engineer is a prompt engineer. Everyone is integrating AI products into their product lines," he noted.

Arize now serves major enterprises such as Uber, Klaviyo, and Tripadvisor, and also offers an open-source product, Arize Phoenix, which boasts over two million monthly downloads. The Berkeley, California-based company recently secured a $70 million Series C funding round led by Adams Street Partners, with participation from M12, SineWave Ventures, OMERS Ventures, and strategic investors like Datadog and PagerDuty. This brings Arize's total funding to over $130 million.

The new funds will be used to enhance Arize's core product and expand into growing AI segments, including voice and AI agents. Dhinakaran, while acknowledging that their open-source product might be their biggest competitor, emphasized their commitment to further developing it, saying, "Our open source Phoenix has just been growing, it’s been growing massively, and so I think we love that. We love open source."

The AI observability and evaluation market is becoming increasingly competitive. Arize differentiates itself by offering both pre- and post-launch evaluations across various AI applications. However, companies like Galileo, with $68 million in venture funding, and Patronus AI, with $20 million, offer similar services.

Lopatecki highlighted the challenge of building the necessary infrastructure for AI observability, suggesting that this is why major players like Microsoft and Datadog are investing in Arize. He anticipates a rapidly growing market, with both small startups and large corporations vying for a piece of the pie. "I think people also now see how big this market can be," he added.

This article has been updated to accurately reflect the founding year of Arize.