Report: AI Agents Increase Observability Costs

Cisco has released its Splunk State of Observability 2026 report, revealing that companies face growing costs and operational challenges as they deploy AI agents without sufficient monitoring capabilities.

The report found that 57% of organizations spend more on telemetry, the data used to monitor IT systems, than on the infrastructure being monitored. Additionally, 55% of respondents said they can’t adequately control their AI costs, while 60% reported that unexpected billing spikes create greater operational difficulties than consistently high expenses.

Conducted by Oxford Economics, the survey polled 2,050 IT operations and engineering professionals across 10 countries between February and March 2026.

Overall, the findings reveal a real challenge for companies adopting agentic AI. The technology needs extensive monitoring, yet the monitoring itself creates major costs. Companies must develop monitoring capabilities that track performance and spending without allowing telemetry costs to undermine the benefits of AI.

More Monitoring Data

A key contributor to these costs is the large volume of data produced by AI agents. According to Kamal Hathi, senior VP and general manager of Splunk at Cisco, production AI agents generate substantial telemetry data as they interact with AI models, external tools, authentication systems and other agents. This additional monitoring data can boost observability costs.

But AI is not the only factor driving monitoring expenses. The survey found that 31% of respondents identified excessive log data as their largest observability cost, followed by high-cardinality metrics at 27%, and trace volumes at 21%.

To cut spending, 38% of companies have limited their telemetry collection or retention. The upside to this strategy is that it can reduce storage costs, but it deprives network teams of historical data needed to diagnose intermittent performance issues.

AI Monitoring Capabilities Lag Behind Adoption

The survey found significant weaknesses in enterprises’ ability to monitor AI workloads. Although 87% of respondents consider AI monitoring a priority, 56% cannot measure essential performance indicators like LLM response quality. Another 49% reported that their existing observability tools lack sufficient AI monitoring features.

Responsibility for monitoring AI also remains unclear at many organizations, with 38% reporting that the task has no clearly designated owner. These limitations are particularly worrisome given that 83% of respondents worry about unintended actions by AI agents.

Unlike conventional software, AI agents independently initiate tasks and interact with business systems. So monitoring requires visibility into both technical performance and the business results of agent decisions.

Security Teams Take Larger Role

The deployment of AI agents is also changing how IT operations and security departments work together. Cisco found that 81% of respondents reported greater collaboration between observability and security teams as a result of AI monitoring.

Plus, the percentage of organizations able to connect performance problems to security-related causes rose from 65% in 2025 to 73% in 2026. Also positive: 84% reported sharing data between security and observability teams, compared with 74% previously.

Cisco’s research identified a group of organizations with more advanced observability practices that reported stronger business results. Among this group, 91% could accurately identify AI costs by workload, compared with 69% of other respondents.

Additionally, 54% of the advanced organizations reported at least $5 million in annual revenue associated with observability, compared with 28% of other organizations. However, these figures are self-reported and don’t establish that observability investments directly caused the revenue gains.