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.

The accuracy problem with agentic AI

Greifeneder walked the audience through the math. A single LLM agent running at 95% accuracy ...