Safe Experimentation and Human-in-the-Loop

Perform 2026 felt like a turning point for Dynatrace, and when Steve Tack joined me for his fourth appearance on the show, it was clear this was not business as usual.

We began with a little Perform nostalgia, from Dave Anderson's unforgettable "Full Stack Baby" moment to the debut of AI Rick on the keynote stage. But the humor quickly gave way to substance. Because beneath the spectacle, Dynatrace introduced something that signals a broader shift in observability: Dynatrace Intelligence.

Steve was candid about the problem they set out to solve. Too much focus on ingesting data. Too much time spent stitching tools together. Too many dashboards. Too many alerts. The real opportunity, he argued, is turning telemetry into trusted, automated action. And that means blending deterministic AI with agentic systems in a way enterprises can actually trust.

We unpacked what that looks like in practice. From United Airlines using a digital cockpit to improve operational performance, to TELUS and Vodafone demonstrating measurable ROI on stage, the emphasis at Perform was firmly on production outcomes rather than pilot projects. As Steve put it, the industry has spent long enough in "pilot purgatory." The next phase demands real-world deployment and real return.

A big part of that confidence comes from the foundations Dynatrace has laid with Grail and Smartscape. By combining unified telemetry in its data lakehouse with real-time topology mapping and causal AI, Dynatrace is positioning itself as the engine behind explainable, trustworthy automation. When hyperscaler agents from AWS, Azure, or Google Cloud call Dynatrace Intelligence, they are expected to receive answers grounded in causal context rather than probabilistic guesswork.

We also explored what this means for developers, who often carry the burden of alert fatigue and fragmented tooling. New integrations into VS Code, Slack, Atlassian, and ServiceNow aim to bring observability directly into the developer workflow. The goal is simple in theory and complex in execution: keep engineers in their flow, reduce toil, and amplify human decision-making rather than replace it.

Of course, autonomy raises questions about risk. Steve acknowledged that for now, humans remain firmly in the loop, with most agentic interactions still requiring checkpoints. But as trust grows, so will the willingness to let systems self-optimize, self-heal, and remediate issues automatically.

We closed by zooming out. In a market saturated with AI claims, Steve encouraged listeners to bet on change rather than cling to the status quo. There will be hype. There will be agent washing. But there is also real value emerging for those prepared to experiment, learn, and scale responsibly.

If you want to understand where AI observability is heading, and how deterministic and agentic intelligence can coexist inside enterprise operations, this episode offers a grounded, practical perspective straight from the Perform show floor.