Dynatrace、AI運用の難所を特定する新エージェントを発表

Observability platform company Dynatrace announced a set of advancements to its Dynatrace Intelligence service on Monday that aim to take the autonomous SRE (site reliability engineer) out of its probabilistic approach of the past. The goal is to move forward to now work with more deterministic real-time context and control.

New extensions are designed to automatically resolve software infrastructure operations incidents and prevent service disruptions, while still maintaining human oversight and governance.

Dynatrace Intelligence was initially launched at Perform, the company’s flagship user conference, in January of this year. Dynatrace is now adding new autonomous agents for incident triage and remediation, as well as no-code custom agent creation capabilities. The platform is also expanding its integrations to deliver autonomous observability functions in commonly used tools and workflows.

Does this mean we can start celebrating the year of the autonomous SRE (in 2027 perhaps) and cross the chasm to true unattended autonomy?

Steve Tack, chief product officer at Dynatrace, tells The New Stack that “it’s not a binary choice between human and platform” i.e. human-led and agent-led operations continue to coexist. He encourages us to realize that the real goal is to “move to more autonomous operations over time” as confidence in the results grows.

A crawl-walk-run approach

“With customers that have adopted the autonomous SRE approach (as well as us), we often see a crawl-walk-run approach,” Tack says. “First, there is trust in our deterministic and causal AI precisely pinpointing the root cause; second comes remediation automation on top of that (which can be based on both our causal as well as predictive capabilities and still requires a human-in-the-loop), the step towards human-on-the-loop and truly autonomous actions is based on the confidence achieved in initial steps.”

“With customers that have adopted the autonomous SRE approach (as well as us), we often see a crawl-walk-run approach; the step towards human-on-the-loop and truly autonomous actions is based on the confidence achieved in initial steps.”

Tack explains that most teams start with humans-in-the-loop, validating what Dynatrace already knows and does with precision, before they let the platform act on its own.

“The measure of success isn’t how productive your engineers are – it’s the percentage of incidents that never need a human at all,” Tack clarifies. “Dynatrace provides suggestions and pointers, helping the user identify what should be automated based on triaging. The level of autonomy is up to the user to decide.”

AI with deterministic, real-time understanding

Detailing its current round of product updates, Tack and team have noted that Dynatrace has combined agentic AI with deterministic, real-time understanding of complex environments. It’s all about what the company positions as “AI that acts on facts, not guesses” today.

Looking at the star of the show, the autonomous SRE agent triggers autonomously on newly detected problems to determine whether the issue is are part of an existing incident investigation. If confirmed, the agent enriches the investigation with additional insights and updates the detected problem with a reference to the ongoing investigation.

Staying in the SRE agent family, a cloud SRE agent coordinates remediation activities and integrates with agents across AWS, Microsoft Azure, and Google Cloud environments. It does so before centralizing findings to provide a single auditable record for autonomous operations.

Three other key product updates are also here. Agent Builder creates and deploys custom AI agents without code, extending autonomous operations to workflows unique to their environments; an enhanced iteration of Dynatrace Assist brings natural-language investigation and agent-ready workflows to users; and deeper connection to cloud hyperscalers sits alongside new integrations with ServiceNow, Atlassian, and PagerDuty.

Pessimistic about anachronistic probabilistic services

Dynatrace noted that its move beyond observability that relies on probabilistic outputs is achieved by Dynatrace Intelligence grounding every action in deterministic, real-time system understanding. Every action is rooted in environment-specific context and designed to be transparent, auditable, and governed.

“The gap between AI-generated insight and safe, governed execution is one of the biggest concerns; customers need deterministic, real-time context with automation and auditability to drive trusted and reliable outcomes.”

“Enterprises investing in AI-driven observability have an opportunity to turn data into intelligence that translates into trusted, autonomous action,” said Stephen Elliot, group VP at analyst house IDC. “The gap between AI-generated insight and safe, governed execution is one of the biggest concerns; customers need deterministic, real-time context with automation and auditability to drive trusted and reliable outcomes.”

Cloud SRE Agent, Enhanced Dynatrace Assist, and the expanded integration ecosystem are available to SaaS customers on DPS today. Autonomous SRE Agent and Agent Builder are expected to be available in August 2026.

2027: The year of the autonomous SRE?

Looking ahead, we can expect 2027 to be the Chinese year of the goat, the FIFA Women’s World Cup and perhaps the year of the autonomous SRE (site reliability engineer).

While only two out of those three are a certainty, with platforms such as Dynatrace, Microsoft, Datadog and Komodor all shipping policy-bound AI agents to drive autonomous SRE functions, next year could see us cross the chasm to true unattended autonomy.

The post Dynatrace’s new agents can reveal the single hardest part of AI operations appeared first on The New Stack.