Dynatrace Intelligence Fuses Agentic AI with Determinism for Autonomous Operations

Dynatrace Perform 2026 - why agentic AI only works when determinism comes first

On the ground at Dynatrace Perform 2026, you couldn't accuse the vendor of being dull with the presentation. The morning keynote opened with CEO Rick McConnell attempting to hand over to an AI-generated version of himself. AI-Rick variations appeared on the Venetian Ballroom's massive screens - Rock Rick, Tech Rick and Relaxed Rick - before he dismissed them. The stunt made a serious point about trust in AI systems. When CTO Bernd Greifeneder took the stage later to unveil Dynatrace Intelligence, he demonstrated why most agentic AI implementations fail in production. You have an LLM (Large Language Model) with 95% accuracy, and then you do multiple agentic calls - let's say 10 in a row - the error rate accumulates, and the end result is 60%. This is not acceptable. Greifeneder walked the audience through the math. A single LLM agent running at 95% accuracy sounds acceptable - until you chain multiple sequential calls together solving a complex incident. By the tenth step, success rates can drop to around 60%. You cannot take a petabyte of logs and load into an agent, because it bursts the context window. And the more data is in there, the higher risk of hallucinations. He discussed the need to maximize determinism before any LLM sees the problem: The better structured the information is, the more context you have, the more causal information you provide to an AI, the better the outcome is. Modern observability reveals why this matters - a degraded checkout flow might trace back to database contention from unrelated analytics queries. When AI agents must traverse these causal chains without deterministic grounding, small errors compound quickly across each inference step. The architecture addresses this with three deterministic AI agents establishing factual grounding before generative AI enters the workflow. The root cause agent analyses millions of causal dependencies using Dynatrace's Smartscape topology graph. An analytics engine transforms exabytes of data in the company's Grail lakehouse into context-rich information optimized for AI consumption. A forecasting agent scales predictive analytics across millions of metrics simultaneously. Benchmark results show 12 times higher success rates, three times faster resolution, and half the token costs compared to LLM-only approaches. Greifeneder added: And here's the kicker. The bigger, the more complex your environment is, the better those numbers get. When Ramiro Zavala, Head of IT Operations, Observability and Quality at United Airlines, walked onto the stage with Dynatrace CMO Laura Heisman, she reminded him of their last meeting - an early morning lobby encounter in Texas last October when he was already deep in crisis calls at 2am. Zavala recalled: This one was definitely different than how we would have handled it two years ago. Two years ago, diagnosing what was happening during a major incident required "upwards of 250 people". United operates more than 2,000 application services running 24/7. A single boarding pass triggers up to 500 unique services. The airline's infrastructure spans mainframes, on-premises systems, and modern cloud applications -- all still heavily reliant on legacy technology from United's 100-year history. Zavala's team consolidated fragmented monitoring tools onto Dynatrace, moving approximately 800 applications in nine months through "developer Observability days" and "Dynatrace blitz months." The results: two best years on record back to back, number one in the industry for on-time departures, customer satisfaction scores up 2.6 points last year. Zavala credits the shift from signal-watching to business outcome focus. His team built internal "digital control towers" that connect technical metrics directly to business expectations. He observed: Having a conversation about maybe a kiosk not printing the right number of bag tags is a lot different than a kiosk not being available. It brings us to the table with our business partners to have that kind of conversation. He added that the next chapter is agentic automation integrated with ServiceNow workflows. Greifeneder's big reveal came near the end of his keynote session: Dynatrace Intelligence as "a fusion of deterministic AI with agentic AI, deeply built into the Dynatrace platform natively." The system includes an operator agent handling planning and orchestration, plus domain-specialized agents targeting SRE, development, and security teams. When a high-impact vulnerability surfaces, the system assesses exposure automatically, correlates evidence across logs and traces, examines runtime behavior, and generates prioritized response plans flowing directly into ServiceNow, GitHub, Atlassian, AWS, Azure, Google Cloud, and other enterprise platforms. The Model Context Protocol (MCP) implementation demonstrated earlier connects this to developer workflows. Developers using Visual Studio Code with GitHub Copilot can query slow services and the AI assistant - designed to be production-aware through MCP integration - accesses active problems, identifies slow queries, and surfaces root causes like database contention from background analytics processes. The intended outcome is that developers implement the fix, validate it with observability metrics, and confirm resolution without leaving their IDE. The architecture emphasizes controlled autonomy so that enterprises can progress through phases: AI-driven insights and recommendations, supervised automation with human oversight, then fully autonomous operations with policy-based guardrails. This approach is designed to allow teams to build confidence incrementally rather than forcing wholesale operational changes. Chief Product Officer Steve Tack brought ServiceNow EVP Pablo Stern onto the stage to discuss the integration, which both executives framed as a question of trust - the confidence enterprises need before allowing automation to act in production. He cited early results from non-profit care organization CareSource: 45% reduction in mean time to resolution, 35% increase in self-healing through closed-loop incident management connecting forensics to automated paging and team coordination. But both executives emphasized the foundation required. Stern said: We often talk about having prescriptive workflows and prescriptive relief because in order to build the trust, you have to start from a foundation of what we ultimately are trying to drive and deliver. Capabilities launching this week include skills for understanding blast radius and driving root cause analysis. Pre-flight checks validate changes before deployment by understanding potential risk and blast radius. Both companies committed to production validation - Dynatrace will implement ServiceNow for internal operations while ServiceNow deploys Dynatrace observability for its digital operations, validating the integration under real operational load before broader customer rollouts. Tack framed a shift from cloud monitoring to cloud operations. "One of the common themes we hear is that we're grounding in data, but we're still starving for action." Moving Smartscape - Dynatrace's real-time dependency graph - into Grail allows topology data to be queried alongside other observability signals, tightening the link between telemetry and operational decisions. The implication is that observability is no longer just about seeing systems, but about shaping when and how action is taken. Enhanced monitoring across AWS, Azure, and Google Cloud reached general availability on AWS, with Azure and Google Cloud in preview. For AI workloads, Dynatrace extends visibility across the full stack, from prompts and models to infrastructure - a response to a new operational reality. As agentic systems generate increasingly unique interactions, the traditional logic of troubleshooting begins to break down. Tack posed a question, leaving the audience with a problem that extends far beyond monitoring: When every agentic interaction can be unique, how do you troubleshoot that? How do you manage it? Mala Pillutla, VP of Regional Sales Management, addressed log management and real user monitoring - two capabilities that historically frustrated enterprise teams. Traditional log management forced bad choices: pay a fortune for siloed data, or sample logs to save money and lose root cause evidence. Dynatrace's approach manages telemetry at ingest, uses AI to translate log lines contextualized with trace data, and provides cost visibility with chargeback perspectives. On real user monitoring, she described how every click, swipe, and scroll tells a story: From a business stakeholder perspective, if I have a critical revenue generating app, I would like to know why a customer clicked on purchase and then didn't complete it. The enhanced Real User Monitoring (RUM) aims to unify user interactions with logs, metrics, and traces in Grail. AI-generated user session narratives explain what happened and why for both IT and business stakeholders. An error inspector app auto-prioritizes errors impacting customer attrition and revenue to, as Pillutla put it, "fix what matters, not what's the loudest." McConnell opened the morning declaring "AI is your team" - amplifying human capabilities rather than replacing them. But he conditioned that promise on reliability: We have to be delivering reliable AI because we have to be certain that the output, the results, are meaningful. The 12x success rate improvement stems from establishing factual grounding through deterministic agents before introducing LLM reasoning. Measurable value concentrates in toil elimination - United's compressed incident resolution is an example of concrete returns beyond theoretical productivity assumptions. The most telling signal from the keynote came from how speakers described the technology - supervised extension rather than autonomous replacement. Dynatrace Intelligence detects anomalies, traces causality across distributed systems, and proposes optimizations. But the decisive moment of whether a change reaches production still belongs to a human engineer. Based on customer evidence, deterministic AI foundations appear essential for enterprises to trust agentic automation with revenue-impacting systems. The question is how quickly organizations can build that trust - and whether they will accept autonomous decisions even when the math says they should.

Dynatrace Intelligence aims to advance autonomous software operations

Dynatrace Intelligence aims to advance autonomous software operations Artificial intelligence observability platform company Dynatrace Inc. today announced a set of platform updates at its Perform Conference in Las Vegas that included the debut of Dynatrace Intelligence, a new agentic operations system designed to bring enterprises closer to reliable, autonomous software operations. The launch marks the next phase in the evolution of the Dynatrace platform by combining deterministic, real-time system intelligence with agentic artificial intelligence that can reason and take action within clearly defined guardrails. The company says the approach is intended to help organizations to move beyond reactive operations towards more autonomous outcomes while also preserving visibility, control and governance across complex environments. The new Dynatrace Intelligence is built on the company's Grail data lake house, Smartscape technology and telemetry data to allow AI agents to operate using precise, environment-specific context rather than relying primarily on probabilistic large language model behavior. The system can deliver trustworthy, explainable insights that AI agents can safely act on, even in high-risk enterprise environments. Dynatrace Intelligence also delivers an ecosystem of AI agents operating at different levels of autonomy. Foundational agents provide core intelligence and reasoning, while task-specific agents apply that intelligence to domains such as site reliability engineering, development, security and business operations. Ecosystem agents extend the platform into third-party tools and services, including integrations with ServiceNow, Amazon Web Services Inc., Microsoft Azure, Google Cloud Platform, Atlassian, GitHub and Red Hat. The system supports assisted workflows, where teams interact via natural language and receive deterministic recommendations and more automated workflows, where actions are executed within predefined guardrails. Dynatrace also today announced expanded cloud operations capabilities that offer deeper cloud-native integrations across AWS, Azure and Google Cloud. The enhanced capabilities are focused on providing enterprises with a more unified view of complex multicloud environments and consolidating visibility into a single control plane to help teams identify issues faster, reduce disruption and turn operational complexity into a business asset. Other updates announced today included a focus on improving the developer experience by evolving observability into an active system of control for AI-native software delivery. New capabilities unify frontend, backend, database, cloud, mobile and AI telemetry into a single developer-facing experience. The company also introduced next-generation Real User Monitoring capabilities that are designed to unify frontend telemetry with backend context. The updated RUM offering is designed to address blind spots emerging as organizations adopt cloud-native architectures and AI-driven applications. Rick McConnell, chief executive officer of Dynatrace, spoke with theCUBE, SiliconANGLE Media's livestreaming studio, in December, where he discussed how Dynatrace is pushing observability into the AI stack.

Dynatrace Pushes The Agentic AI Envelope In Observability With New 'Intelligence' Offering

The launch of Dynatrace Intelligence, along with other announcements at the company's Perform customer and partner event, follows Dynatrace's multi-year pivot to a channel-centric go-to-market strategy. Dynatrace today unveiled Dynatrace Intelligence, the latest generation of the company's observability platform, which fuses deterministic and agentic AI capabilities for more reliable IT operational observability. Dynatrace, which is holding its Perform annual customer conference (with a partner component) in Las Vegas this week, also debuted a number of additional new products and enhancements including Dynatrace Intelligence Agents, which the company said are designed to take action across IT and business operation workflows and drive closed-loop autonomous outcomes. The company also introduced expanded cloud-native capabilities to work with the major cloud platforms, enhanced developer experiences, and the availability of advanced Real User Monitoring capabilities. [Related: Dynatrace Deepens Observability Links To AWS With New SCA] All this comes as Boston-based Dynatrace continues a pivot to the channel that began four years ago under CEO Rick McConnell. Today the company works with about 700 channel partners, including global system integrators and leading solution and strategic service providers: Last year about 80 percent of new sales bookings came "through and with" the company's partner ecosystem, said Jay Snyder (pictured), Dynatrace senior vice president of partners and alliances, in an interview with CRN. Dynatrace is putting increased emphasis on partners who not just sell and implement the company's technology, according to Snyder, but who build observability practices and help customers leverage the data and insights generated by the Dynatrace platform to improve their IT and business operations. (The company is currently exploring working with managed service provider partners, Synder said.) "What my mantra has been -- and it's been fairly consistent for the last year-plus -- is that we're moving from the transaction to the lifecycle-based approach for our customers," Snyder said. "We want our partners to be thinking not about the deal. The deal is just the starting box. And then it's how are we going to manage this customer over time, collectively, to deliver the most success," he said. Partners also bring a wealth of vertical industry expertise to the Dynatrace platform ("That's a huge value for us," Snyder said). And they help customers integrate the Dynatrace technology with customers' broader IT landscape to automate IT remedial tasks, provide operational data to workflow systems such as ServiceNow (Dynatrace and ServiceNow announced a multiyear strategic collaboration deal in October), and support organizations' cybersecurity systems. "Our goal is to use [partners] as the services extension of our company and or in conjunction with our own services business, where they leverage our technical expertise to augment their own delivery services, and we co-deliver, and that model is starting to work really, really well for us," the channel chief said. Part of that approach is that Dynatrace is now providing partners with tools, frameworks and capabilities from its own service organization. Over the last year and a half Dynatrace and some partners have been signing what Snyder calls "teaming agreements" that go beyond traditional deal registration by spelling out the needs, deliverables and expected outcomes of customer engagements. "It is not just that we are holding ourselves and our partners accountable, but the customer gets the best possible outcome because we work back from the customer," the channel chief said. "It means that we're partnering much more strategically, much earlier in the sales cycle, and we're getting much more value out of the partner. And the partner is getting more value out of Dynatrace." Dynatrace Intelligence Debut Dynatrace describes the new Dynatrace Intelligence offering as the latest phase of the evolution of the company's Dynatrace Platform. Amid increasing IT complexity, including sometimes unpredictable AI and agentic systems, Dynatrace Intelligence is designed to help organizations move from reactive responses to IT events, to proactive remedial and preventive actions, and even advance toward autonomous operations across digital ecosystems, Dynatrace said in today's announcement. An AI engine that works with the Dynatrace Platform, Dynatrace Intelligence is designed to observe and optimize dynamic AI workloads. The software combines deterministic intelligence, grounded in real-time causal context, with agentic AI that can "safely reason, decide, and act within defined guardrails," according to the company. Dynatrace Intelligence stores and unifies data in Grail, Dynatrace's unified data lakehouse system. That data is continuously and automatically enriched by Smartscape, the company's real-time, automated topology mapping technology that visualizes dependencies across an application stack. Also unveiled Wednesday was Dynatrace Intelligence Agents, built on Dynatrace Intelligence, that take action across workflows and drive autonomous outcomes across IT and business operations, according to the company. Increased Cloud Integrations While Dynatrace offers its own AI agents, Dynatrace Intelligence provides bidirectional integrations with agents from other vendors including ServiceNow, Amazon Web Services, Microsoft Azure, Google Cloud, Atlassian, GitHub, Red Hat and others. Dynatrace also announced expanded cloud-native integrations across AWS, Microsoft Azure and Google Cloud Platform, links that the vendor said provide organizations with a clearer, more unified view across multi-cloud environments. Also new are development enhancements for building agentic and large-language model-drive applications. The innovations, according to Dynatrace, unify front-end, back-end, AI telemetry, database, cloud and mobile systems into a single developer-facing experience built on Grail, Smartscape and Dynatrace Intelligence. "What we're doing is bringing all of that together into one single platform, so you can understand how everything is connected, from the cloud infrastructure to the application to the end user experience and even to the business outcomes," Snyder said. "And they're not just pulling in the raw data, but they feed it into the Dynatrace intelligence layer...which will then use the AI to explain what's happening. This is incredibly powerful and useful for a development team." The channel chief said the new development capabilities will be especially welcomed by Dynatrace's system integrator partners. Dynatrace also announced the launch of next-generation Real User Monitoring (RUM) capabilities, combining front-end telemetry with back-end context, to empower teams to better understand and optimize user experiences.