AI agents are giving IT teams more to monitor, from the quality of their responses to the cost of running them. But many teams struggle to measure AI performance or control its costs, while the data collected to monitor technology systems can itself cost more than the infrastructure being watched, according to a new survey.
Cisco-owned Splunk, an observability software provider, reported the gaps in its report “State of Observability 2026: Restoring Predictability in the AI Era.” Oxford Economics surveyed 2,050 IT operations and engineering professionals in 10 countries between February and March 2026, including developers, engineers and technology executives.
Monitoring AI was an organizational priority for 87% of respondents. Yet 54% said they had no clear business outcomes tied to that monitoring, and 56% could not measure AI workload metrics such as the quality of language-model responses. Although 85% considered visibility into AI hallucinations important, only 58% said they were effective at monitoring them.
Monitoring data adds to the bill
Cost control was another weak point: 55% said they could not accurately control AI spending, and 60% said unexpected billing spikes caused more operational strain than consistently high costs.
A separate finding concerned the monitoring systems themselves. 57% said their telemetry costs exceeded the infrastructure costs of the systems being monitored. Telemetry includes logs, metrics and traces that help IT teams understand how systems perform and diagnose problems. The figure covers observability costs broadly; it does not establish that AI monitoring alone exceeded AI infrastructure spending.
Splunk also identified a more familiar source of expense behind the monitoring bill. In a separate discussion of the findings, Splunk executive Kamal Hathi said excessively detailed logs were the most frequently cited observability cost driver, while only 6% named AI infrastructure as their biggest observability cost driver. Neither Cisco’s announcement nor Hathi’s blog quantified how much of that telemetry spending was attributable to AI agents.
Advanced teams report better cost visibility
Splunk classified some respondents as advanced observability users based on practices such as using AI agents to resolve issues and connecting monitoring data to business planning. Within that group, 91% said they could assign AI costs accurately to particular agents, models or workloads, compared with 69% of other respondents. The comparison does not establish that those practices caused the difference.
The survey also found that 81% of respondents said monitoring AI workloads had increased collaboration between observability and security teams. The share reporting that they could trace performance problems to security causes reached 73%, compared with 65% in the previous year’s survey.
The findings relied on respondents’ descriptions of their organizations, rather than audited invoices or performance tests.