From dashboards to decisions: AI incident detection grows up
AI incident detection in modern enterprises means using observability platform AI and autonomous SRE agents to continuously analyze telemetry, identify production problems, propose or execute incident remediation automation workflows, and shorten the time it takes humans to understand and fix service disruptions, while keeping engineers in control of automated actions instead of replacing them outright. The latest moves from Dynatrace and Expedia show a clear shift: the future of operations is not more dashboards, but more decisions made by AI agents that act on reliable context rather than probabilistic guesswork. Dynatrace is evolving its Intelligence service so autonomous SRE agents work with deterministic, real-time understanding of complex environments, creating AI that “acts on facts, not guesses”. Expedia, meanwhile, is investing in AI-assisted observability with its Service Telemetry Analyzer to accelerate how engineers investigate incidents without surrendering production control.
Expedia’s STAR: AI observability without surrendering autonomy
Expedia’s Service Telemetry Analyzer (STAR) is a pointed rejection of fully autonomous AI incident detection in favor of disciplined, human-centered observability platform AI. STAR ingests standardized infrastructure metrics from Kubernetes services and JVM applications—throughput, latency, error rates, resource utilization, container restarts, and JVM memory and garbage collection—and runs them through predefined diagnostic workflows powered by large language models. The result is structured root cause assessments and recommended next steps that cut the time engineers spend hunting for the source of degradation, while keeping humans responsible for validating findings and deciding what to do next. That matters: AI is used to minimize “time to know” and “time to recover,” not to decide on its own what production changes to push. Technically, STAR favors deterministic prompt chaining over autonomous tools, avoiding function calling, RAG, memory, or open-ended tool use for the sake of consistent, audit-ready analyses.
Dynatrace’s autonomous SRE agents: automation that acts on facts
Where Expedia stays firmly human-driven, Dynatrace is betting that enterprises are ready for AI incident detection that can act, not just advise. The company has announced major advancements to its Intelligence service, adding autonomous SRE agents for incident triage and remediation automation, along with no-code agent creation capabilities. The autonomous SRE agent now triggers on newly detected problems, determining whether they belong to an existing investigation and, if so, enriching it with further insights and linking the problem back to that investigation. A Cloud SRE Agent coordinates remediation activities across AWS, Microsoft Azure, and Google Cloud, keeping a single auditable record for autonomous operations. This is the heart of Dynatrace’s argument: by grounding every action in deterministic, real-time context, its observability platform AI can move beyond probabilistic outputs and unreliable automation, instead offering incident remediation automation that is transparent, governed, and rooted in environment-specific facts.
Human-in-the-loop: preventing runaway autonomy in production
The most important design choice in these systems is not which LLM or metric they use—it is the insistence on human oversight. Dynatrace’s new extensions are explicitly designed to automatically resolve operations incidents and prevent disruptions while maintaining human governance. Teams start with humans-in-the-loop, validating what autonomous SRE agents know and do, and only move towards human-on-the-loop and unattended actions once they build confidence in precise, deterministic root cause detection. One operations leader notes that automation grounded in real-time context reduces manual effort, allowing teams to focus on higher-value work and improving operational outcomes. Expedia takes an even more conservative stance: STAR is AI-assisted, not agentic, and keeps engineers responsible for validation and decision-making. The message is clear: real-time automation without this kind of guardrail is a liability, and enterprises will only tolerate autonomous SRE agents when they are demonstrably under policy, audit, and human control.
What this shift means for enterprise operations teams
The immediate impact of these platforms is practical, not cosmetic. By combining telemetry with LLMs in deterministic workflows, Expedia’s STAR cuts the manual investigation overhead that used to dominate incident response. Dynatrace’s autonomous agents go further, acting on that understanding to automate triage and multi-cloud remediation coordination, reducing the number of incidents that ever require human intervention. According to Dynatrace’s chief product officer, the real measure of success is “the percentage of incidents that never need a human at all,” not how busy engineers appear. That is a provocative stance—but a necessary one. As incident volume rises and systems grow more complex, doing more with the same number of engineers means pushing routine detection and remediation into AI incident detection and observability platform AI. The next wave of enhancements—custom agents, richer dependency data, chaos engineering analysis, and conversational interfaces—will only accelerate this trend. The teams that embrace governed autonomy early are likely to see the biggest gains in mean time to resolution.
