Dynatrace, Inc. Goldman Sachs Communacopia + Technology Conference 2026

Thanks for having us. We are always happy to be here.

I probably need to move to new pants. Yeah. Very well. With my exploding Perrier.

Yeah. I think we'll survive.

Well, there is no doubt that the observability category is evolving in a substantial way in the age of AI. In an AI-first world, we would submit that the die has been cast, that observability is more critical each day. I remember 6 months ago, 9 months ago, when we were in the midst of SaaSpocalypse, when it was deemed that the vast majority of software could be rewritten. It has evolved since then. I think we've evolved to the point where there are clearly categories that are winners and maybe some other categories that will struggle more in an LLM-driven world. I would submit to you that observability is one of those categories that is going to win. We are entering a new era. In observability, we have all the traditional software observability that we've been driving today.

This is a couple of billion dollars of ARR plus. It's growing at 17% as our core business, and that continues to evolve to a world where customers want automation. They want autonomous operations. The goal in this traditional business is that you have software that essentially runs itself, that corrects itself, that auto-remediates, and that can reduce the human load required to manage it. Similarly, though, you have this new expanded category, which we're thinking of as AI observability. AI observability is the observability of AI workloads and of agentic models. This is a market that we believe is going to expand to in the range of a $10 billion market by the end of the decade, growing well more than 50% per year.

We've got large, traditional, $80 billion-plus traditional observability market growing in the mid-teens, supplemented by a brand-new category in AI observability that is even more critical because you need more observability to oversee AI workloads, not less. What you need is context to be able to do that, and that is what observability systems provide in real time by analyzing billions of interconnected data points. That is a huge differentiator from LLM models and what they would be doing.

It's a great question, Matt. The biggest delta that I see, and I always have believed that I'm in a privileged position of being able to meet with CIOs, CXOs all over the planet of Global 500, Global 2000, even Global 1500 organizations. Near as I could tell, the inhibitor to broader-based deployment of AI is confidence. What I mean by that is once you start relying upon LLMs to provide data to end users from your mobile app, from your website, from whatever it might be of the company that would be our customer, the challenge is you better make sure that that information coming out of that LLM is right.

Because if a bank, for example, has a prompt in a chatbot sitting in their mobile app and somebody says, "Well, where should I invest $10,000?" The answer comes back, "Well, the highest return over the last 90 days was to invest in Bitcoin, so why don't you put all of your money in Bitcoin?" Probably not the right answer, depending upon the characteristics of that investor. The calibration of the input coming from the LLMs to deliver to end users is really critical, and it is that confidence level that I think is required to get over the next hump. This is precisely what the domain of AI observability is. AI observability is around LLM experimentation, LLM observability. It's around AI development life cycles and workloads.

The result of this, or the intent of this, is to ensure that you're answering an incremental question from the traditional question of observability. The traditional question that we're trying to answer in observability, this couple billion-dollar business that I talked about of Dynatrace's, is is it running? Are your workloads running the way that you would expect? Are they optimized? Are they delivering? Are they up? What is the availability time? These are the elements, the domains of traditional observability. Once you move into an AI world, you have to ask an additional question. Those AI workloads also have to be running, so you start with the same question, but you then expand to another question. The next question is really around is it accurate?

Meaning that are the LLMs delivering accurate information that can be relied upon to be delivered to end users so that you as an organization have confidence that the end user taking action on that can actually make discernible progress. The third question that is associated with those AI workloads is, are my models working right? Are my agents behaving the way that you would expect them to behave? The point is, in a traditional observability sense, is it resilient, is it working, is a pretty good start. In AI workloads, you need to supplement with a couple of additional questions related to the accuracy of the data flow. That's where, A, it's getting really exciting for observability, but B, observability is becoming completely mission-critical in delivering AI workloads with confidence and successfully.