- Datadog released “Bits Investigation” on July 1, 2026, and Splunk reached general availability with “AI SRE” in June 2026, both targeting autonomous incident response and root cause analysis.
- These agentic observability platforms aim to reduce Mean Time To Resolution by autonomously correlating telemetry and suggesting remediation steps rather than leaving that work to engineers.
Within weeks of each other, Datadog and Splunk both shipped agentic incident response tools, and the timing is not coincidental. Datadog released “Bits Investigation” on July 1, 2026, positioning it as an AI responder that works alongside engineers during live incidents. Splunk hit general availability with “AI SRE” in June 2026, an agent designed to detect, troubleshoot and recommend remediation without waiting for a human to start the diagnosis.
Traditional observability tools collect enormous volumes of telemetry, logs, metrics and traces, but leave engineers to do the hard work of correlation and diagnosis. That manual process breaks down fast in modern distributed systems: too many signals, too many services, too much alert fatigue. According to Datadog, Bits Investigation analyzes incident context alongside human teams, reasons over latency graphs shared in chat, pulls in summaries from incident bridge calls and cross-references runbooks in linked repositories to build hypotheses from available evidence. The goal is to shrink the initial investigation window, not eliminate the engineer.
Visibility into what AI agents are doing, whether they are behaving as intended and operating within established guardrails, is becoming a practical concern for any team running agents in production. A customer service AI agent authorised to issue a $600 refund might instead process $1,500 due to a reasoning error, the kind of failure that conventional monitoring catches too late, if at all.
Autonomous Investigation and Predictive Action
The difference between passive monitoring and agentic observability is causal context. Older tools tell you something broke. Agentic systems try to tell you why, and in what order. If a microservice starts showing high latency, an agentic system can trace requests across multiple services, identify a database bottleneck, correlate it with a recent schema change and recommend reverting that change, ideally before a human engineer has finished reading the initial alert.
Dynatrace demonstrated this approach at Perform 2026 in January with Dynatrace Intelligence, which fuses deterministic AI with coordinated AI agents built on its Grail data lakehouse and Smartscape causal topology. The system generates explainable outputs and operates within defined guardrails, meaning it can suggest or initiate remediation steps, rolling back a faulty deployment, scaling up resources, without going off-script. Explainability matters here: an agent that takes autonomous action but cannot show its reasoning is a liability, not an asset. That principle applies equally to the kind of enterprise agent testing teams are now building into pre-release workflows.
Rather than embedding observability into the infrastructure layer, Galileo focuses on agent observability and guardrails at the application level, a different bet on where the monitoring gap actually sits.
Gartner predicts that by 2028, a significant share of organisations deploying AI will adopt dedicated AI observability tools to monitor model drift, bias and LLM behaviour. Gartner also reportedly put approximately $234 billion in enterprise application software spend at risk from agentic AI displacement, a figure that helps explain why vendors are racing to own this layer before the market consolidates. For more on AI agents and automation tools, visit our AI Agents section.
