AI agents are driving up telemetry costs and complicating network management as enterprises struggle to monitor the technology and control unexpected expenses, according to new research from Cisco Splunk.
The company’s State of Observability 2026 report found that 57% of organizations have observability costs that exceed the cost of the infrastructure being monitored. Another 60% said unexpected billing spikes create more operational strain than consistently high costs, and 55% cannot accurately control their AI expenses.
Cisco Splunk surveyed 2,050 IT operations and engineering professionals across 10 countries to understand how AI agents are affecting observability practices and costs.
“A production agent emits 4-5x more data as telemetry,” said Kamal Hathi, senior vice president and general manager of Splunk at Cisco, in an interview with Network World.
Beyond traditional application traces, agents generate telemetry about workflows, tool calls, retrieved context, model interactions, and authentication. As agents interact with other agents, the volume of telemetry generated during multi-agent sessions can increase exponentially, Hathi explained.
“AI agents are both a source and consumer of machine data,” said Greg Leffler, director of developer evangelism at Splunk. “So, this two-fold demand creates an influx of data volume fast, contributing to unpredictable costs.”
Rising telemetry costs create network blind spots
AI agents are adding to an existing telemetry cost problem, as organizations continue to collect large volumes of data that may not be necessary, Leffler said.
Just 6% of respondents identified AI infrastructure as their biggest observability cost driver. Noisy logs topped the list at 31%, followed by high-cardinality metrics at 27% and trace volumes at 21%. Leffler acknowledged that the research did not quantify how much of the cost increase comes from AI agents compared with traditional data collection practices.
The report found that 38% of organizations have reduced telemetry coverage or retention to control costs, potentially creating network troubleshooting blind spots.
“Historical and one-off troubleshooting is often what is sacrificed when data coverage/retention is cut, and this creates those blind spots for network teams,” Leffler said. He added that shorter retention periods limit historical investigations, while reduced sampling increases the risk of missing intermittent network problems.
AI agents introduce new observability challenges
Beyond telemetry costs, organizations must monitor AI agents to ensure they operate correctly and safely within established guidelines.
While 87% of respondents said monitoring AI is a priority, Cisco Splunk’s research identified several challenges:
- 56% cannot measure AI workload metrics such as large language model response quality.
- 49% said their observability tools lack adequate AI monitoring capabilities.
- 38% said their organizations have not clearly defined responsibility for monitoring AI.
“Tools alone can’t determine whether an output is good; that requires an agreement on what ‘quality’ means for each use case, and the skills to turn that definition into accurate evaluations,” Hathi said.
Network operations teams have an important role because agents generate significant network traffic, Leffler said. Teams must monitor traditional metrics such as latency and packet loss while working with security teams to distinguish legitimate agent activity from malicious traffic. For example, an AI agent repeatedly calling an API could generate traffic resembling an attack.
The research also found that 81% of respondents said monitoring AI workloads has increased collaboration between observability and security teams.
Controlling costs and improving visibility
To manage telemetry costs and improve agent visibility, Leffler recommends deploying the vendor-neutral OpenTelemetry Collector.
“With the OpenTelemetry Collector, which is a tech-agnostic tool to process and export telemetry data, you can use processors to both manage costs and help tag traffic sources to provide increased visibility into agents and their infrastructure,” Leffler said.
For network teams, the challenge is no longer just managing the growing volume of telemetry generated by AI agents. It’s determining how much data they need to maintain visibility without allowing observability costs to spiral out of control.