Dynatrace’s new agents can reveal the single hardest part of AI operations

Dynatrace rolled out a fresh batch of updates to its Dynatrace Intelligence platform on Monday, and the pitch this time is less about flashy automation and more about trust. The company wants its autonomous SRE agents to stop guessing and start acting on hard, real-time facts about a system's state, rather than probabilistic best guesses. That's a meaningful shift for a category that's spent years leaning on AI outputs that sound confident but aren't always grounded in what's actually happening in the environment.

Steve Tack, Dynatrace's chief product officer, described a crawl-walk-run path that customers (and Dynatrace itself) tend to follow. First, teams learn to trust the platform's ability to pinpoint root causes using deterministic and causal AI. Then comes remediation automation, still with a human checking the work. Only after that confidence builds do teams move toward what Tack calls human-on-the-loop operation, where the system acts and people supervise rather than approve every step. Tack was blunt about what should count as progress: not how busy engineers are, but how many incidents get resolved without ever needing a human to weigh in.

The product lineup backs up that philosophy. A new autonomous SRE agent watches for newly detected problems and checks whether they're already part of an ongoing investigation, then folds in extra context if so. A companion cloud SRE agent handles remediation coordination across AWS, Azure, and Google Cloud, pulling everything into a single auditable record. Dynatrace also launched Agent Builder, a no-code tool for spinning up custom agents tailored to a team's own workflows, plus an upgraded version of Dynatrace Assist for natural-language investigations, and new integrations with ServiceNow, Atlassian, and PagerDuty on top of existing hyperscaler connections.

Stephen Elliot of IDC put his finger on the real tension here: there's a gap between AI generating an insight and an organization feeling safe enough to let that insight trigger action without a person double-checking it. That gap is exactly what Dynatrace is trying to close by grounding every agent decision in specific, auditable, governed context instead of a probability score. Cloud SRE Agent, the enhanced Assist, and the wider integration set are live now for SaaS customers on DPS; the Autonomous SRE Agent and Agent Builder aren't due until August 2026, so the harder claims about unattended autonomy remain, for now, a promise rather than a shipped reality.

Dynatrace isn't alone in chasing this. Microsoft, Datadog, and Komodor are all building similar policy-bound agents aimed at the same destination — SRE work that runs largely without a human standing over it. Whether 2027 actually becomes some kind of turning point is anyone's guess, but the direction every vendor is pointing in right now is unmistakably the same one.