What happened

Cisco Splunk surveyed 2,050 IT operations and engineering professionals across ten countries. The State of Observability 2026 report found that 57% of organisations see observability costs exceed the cost of the infrastructure they monitor. Sixty percent said unexpected billing spikes add more operational strain than consistently high costs, and 55% cannot accurately control their AI expenses. A production agent emits 4-5x more data as telemetry, according to Kamal Hathi, senior vice president and general manager of Splunk at Cisco. Agents generate telemetry about workflows, tool calls, retrieved context, model interactions and authentication, and multi-agent sessions can increase data volume exponentially.

Why it matters

The surge in telemetry creates network blind spots. The report shows 38% of organisations have reduced telemetry coverage or retention to curb costs, sacrificing historical and one-off troubleshooting data. Shorter retention periods limit investigations, while reduced sampling raises the risk of missing intermittent problems. Beyond cost, 87% of respondents said monitoring AI is a priority, yet 56% cannot measure AI workload metrics such as large language model response quality, and 49% say their observability tools lack adequate AI monitoring capabilities. Network teams must now watch traditional metrics like latency and packet loss while distinguishing legitimate agent traffic from potential attacks.

What this means for enterprise IT teams

Enterprise IT teams face a dual challenge: managing the growing volume of AI-generated telemetry and deciding how much data is needed to maintain visibility without spiralling costs. The research recommends using the vendor-neutral OpenTelemetry Collector to process and export telemetry, applying processors to manage costs and tag traffic sources for better agent visibility. Teams will need tighter collaboration with security groups, as 81% of respondents reported increased cooperation between observability and security teams to handle AI-related traffic. Defining clear responsibilities for AI monitoring and establishing quality metrics for model outputs are also essential steps.

In the words of Kamal Hathi, senior vice president and general manager of Splunk at Cisco: "A production agent emits 4-5x more data as telemetry,"

The bottom line

AI agents are a significant new source of telemetry, driving up costs and complicating network observability. Enterprises must adopt cost-control tools, refine retention policies, and build cross-team processes to monitor both traditional network health and AI agent behaviour. Failure to do so risks blind spots, higher expenses and reduced ability to troubleshoot network issues.

Sources

  • networkworld.com