Splunk’s State of Observability 2026 report, titled “Restoring Predictability to the Agentic Era,” surveyed 2,050 ITOps and engineering professionals across ten countries. Oxford Economics ran the survey between February and March 2026. Its central message is simple: AI agents make software more powerful but also harder to predict, and observability is the key to regaining control.
The Worry Behind the Hype
Traditional systems followed fixed paths. AI agents choose their own. According to the report, 83% of respondents worry about the unintended consequences of AI agents. Yet concern is outpacing capability. While 85% say visibility into AI hallucinations matters, only 58% say they can monitor them effectively. More than half (56%) cannot measure workload metrics such as LLM quality.
Trust Grows, But Carefully
Agents are doing real work. Teams use them frequently or all the time for root cause analysis (62%), improving alert settings (58%), and writing runbooks (57%). Writing new code ranks last: only 6% use agents for it all the time, and 26% never do. The report calls this a “hierarchy of trust,” where teams give agents more freedom as the risk becomes easier to reverse.
The payoff is visible. Some 86% report faster problem detection, and 72% report faster recovery. However, 14% say agents made recovery times worse, which Splunk attributes to the “verification tax” of double-checking agent work.
Who Owns AI Monitoring?
Responsibility is murky. 38% say their organization has no clear ownership of AI monitoring, and 42% say their main challenge is that a different team is responsible. Costs are also slipping out of control: 55% cannot accurately control AI spending.
What “Leaders” Do Differently
The report singles out 206 “leaders” who use agents to fix issues, effectively monitor AI quality, and link telemetry to business results. Their numbers stand out:
- 63% see significant recovery-time gains, versus 40% of others
- 91% clearly define AI monitoring ownership, versus 59%
- 76% can accurately control AI costs, versus 41%
- 54% say observability delivers at least $5 million in annual value, versus 28%
Leaders also collaborate more with security teams (94% versus 80%). Notably, leaders were chosen partly by self-assessment, so readers should treat the comparison with some caution.
The Data Bill Problem
Cost is a recurring theme. Only 6% name AI infrastructure as a top driver of observability costs. Overly noisy logs lead at 31%, followed by high-cardinality metrics (27%). For 57% of organizations, observability bills now exceed the cost of the infrastructure being monitored, and 60% say surprise bills strain them more than high baseline costs. Sampling and smart data routing help, and 38% have had to cut data coverage or retention because of rising costs.
India’s Standout Showing
For Indian readers, the country snapshot is encouraging. Only 46% of Indian respondents cannot measure AI workloads, against 56% globally. Some 74% prioritize AI safety visibility (57% globally), and 73% say telemetry feeds strategic planning (54% globally).
The Takeaway
Splunk closes with four strategies: treat AI quality as a business metric, build trust in phases, bring telemetry into strategic planning, and treat observability as an investment rather than a utility expense. As the report puts it, AI adoption “won’t slow down while observability practices catch up.” Since Splunk sells observability tools, its conclusions favor its own field, but the survey data points to a real and widening gap.
