Cisco’s Splunk has released its State of Observability 2026 report, finding that enterprise adoption of AI agents is outpacing organizations’ ability to monitor their behavior, measure performance and control costs. Based on a survey of 2,050 IT operations and engineering professionals across 10 countries, the research highlights gaps in AI workload monitoring, unpredictable spending and growing overlap between observability and security. While 87% of respondents say monitoring AI is a priority, 54% lack clear business outcomes tied to that effort, underscoring the challenge of turning agentic AI deployments into measurable, manageable business systems.
Splunk Report Highlights the Observability Gap in Enterprise AI
Enterprise AI agents are creating new demands on monitoring systems as organizations deploy software that can interpret information, make decisions and take actions across business applications. Cisco’s Splunk says many companies are struggling to keep pace, with gaps in agent monitoring, cost attribution and security oversight emerging alongside wider adoption.
The company’s sixth annual State of Observability report, titled Restoring Predictability in the AI Era, draws on research by Oxford Economics involving 2,050 IT operations and engineering professionals across 10 countries. Its findings suggest that observability—the collection and analysis of telemetry to understand system performance and behavior—is becoming increasingly important as enterprises move from AI experiments to operational agents.
The challenge extends beyond determining whether an application is online. Organizations must assess whether an agent produces useful responses, behaves as intended, respects security and privacy controls, and delivers value relative to its computational cost.
According to the report, 87% of respondents say monitoring AI is a priority for their organization. Yet 54% do not have clear business outcomes tied to that monitoring, while 56% cannot measure AI workload metrics such as large language model response quality.
The results point to a gap between recognizing the need for AI oversight and establishing the metrics and processes required to manage it.
Agent Observability Remains Difficult to Measure
Traditional application monitoring focuses on signals such as availability, latency, error rates and resource consumption. AI agents introduce additional variables, including response quality, hallucinations, model selection, token usage and whether an agent’s actions achieve the intended business outcome.
Splunk’s research found that 85% of respondents consider visibility into AI hallucinations important, but only 58% say they are effective at monitoring them.
That gap matters when agents are connected to enterprise systems and can trigger actions beyond generating text. An incorrect answer may be relatively easy to correct in a chat interface, but an incorrect action involving customer records, software deployment or operational infrastructure can have broader consequences.
Effective agent observability therefore requires organizations to connect technical telemetry with behavioral and business measures. Teams need to know not only whether an agent completed a task, but whether the result was accurate, safe, authorized and useful.
The survey also suggests that organizations are still working out how to define success. Without agreed business outcomes, monitoring can generate large volumes of technical information without establishing whether AI systems are delivering meaningful improvements.
AI Spending and Telemetry Costs Create New Pressure
Cost predictability is another major concern. As enterprises run agents across multiple models and applications, spending can become difficult to attribute to individual workflows, users or business outcomes.
Splunk reports that 55% of respondents cannot control AI costs, while 60% say unexpected billing spikes create more operational strain than high baseline costs. The findings indicate that financial volatility—not simply the total amount spent—is a significant management issue.
Observability itself can also become expensive. Some 57% of respondents say telemetry costs have exceeded the cost of the infrastructure being monitored. This creates a balancing act: organizations need enough data to investigate problems and understand AI behavior, but collecting and retaining every available signal may undermine the economics of monitoring.
The report says 67% of respondents use a single observability platform, which can provide a unified view of telemetry costs. A consolidated platform may help teams identify which services and workloads generate data, although platform consolidation alone does not guarantee lower costs or better attribution.
For AI systems, a more complete cost model needs to connect usage with the specific agents, models and workflows responsible for it. That can help organizations decide whether an agent’s business value justifies its inference, infrastructure and monitoring expenses.
Observability and Security Teams Are Converging
As AI agents gain access to enterprise systems, monitoring performance and monitoring security increasingly overlap. A slowdown, unexpected action or unusual pattern of requests may reflect a technical problem, a compromised identity, a configuration issue or potentially malicious activity.
Splunk’s survey found that 81% of respondents say monitoring AI workloads has increased collaboration between observability and security teams. The proportion of teams troubleshooting with security colleagues reached 77%, up from 62% in 2025. Data sharing between the teams rose from 74% to 84%.
The report also found that 73% of respondents can accurately trace performance issues to security root causes, compared with 65% in the previous year’s findings.
These trends suggest that organizations are beginning to treat operational telemetry and security signals as related sources of evidence. For agentic systems, that connection can be especially important because agents may interact with identity services, cloud infrastructure, endpoint systems and business applications through automated workflows.
Shared data can help teams investigate incidents more quickly, but effective coordination also requires agreed responsibilities, appropriate access controls and processes for responding to findings. Visibility is a foundation for security, not a substitute for prevention and response measures.
What Observability Leaders Do Differently
Splunk identifies an advanced group of organizations that more frequently use AI agents to remediate issues, monitor factors such as model quality and privacy, and connect telemetry with business decisions.
These organizations report stronger cost attribution: 91% say they can accurately attribute AI costs, compared with 69% of other respondents. They are also 2.4 times more likely to say their observability platforms deliver return on investment significantly above expectations.
Additionally, 54% of the identified leaders say observability has generated at least $5 million in annual business revenue, compared with 28% of non-leaders.
These are survey-reported associations, not proof that a particular observability practice independently causes better financial results. The findings nevertheless suggest that leading organizations are treating observability as a business management capability rather than solely as a technical troubleshooting function.
Regional results also vary. In India, 51% of respondents described their monitoring of AI model quality as very effective, compared with 35% globally. In the United Kingdom, 32% said they use AI agents all the time to diagnose root causes, against a global figure of 20%. Saudi Arabia recorded 98% saying AI had enabled them to focus more on innovation, compared with 84% globally.
Observability Becomes Part of AI Governance
Splunk’s findings point to a broader shift in enterprise AI management. As agents move into production, organizations need a reliable way to measure quality, trace costs, detect unexpected behavior and connect technical findings to business outcomes.
That requires more than adding monitoring to individual AI applications. Enterprises must establish consistent metrics, coordinate security and engineering teams, and determine which signals justify the cost of collecting and retaining them.
The report does not establish a single approach that works for every organization. But its findings indicate that AI observability is becoming a core operational requirement for companies seeking to scale agentic systems without losing control of spending, reliability or security.
Market Landscape
The growth of enterprise AI agents is expanding observability beyond traditional infrastructure and application performance. Organizations increasingly need visibility into model quality, hallucinations, agent behavior, token usage, security events and business outcomes.
Splunk’s report highlights three connected challenges: gaps in AI monitoring, unpredictable AI and telemetry costs, and closer collaboration between observability and security teams. Vendors across observability, cloud monitoring and security analytics are positioned to address parts of this market.
The competitive differentiator will be the ability to connect technical signals with actionable insights and measurable business value. Survey findings indicate growing demand, but monitoring effectiveness, cost savings and return on investment will depend on implementation and workload.
Top Insights
- Monitoring is a priority: 87% of respondents prioritize AI monitoring, but 54% lack clear business outcomes tied to their efforts.
- AI quality remains difficult to measure: 56% cannot measure workload metrics such as LLM response quality, despite widespread concern about hallucinations.
- Costs are unpredictable: 55% struggle to control AI costs, while 60% say unexpected billing spikes create more operational strain than baseline expenses.
- Security and observability are converging: 81% report increased collaboration between observability and security teams as AI workloads expand.
- Leaders connect telemetry to value: Advanced organizations report stronger AI cost attribution and greater-than-expected observability ROI more frequently than other respondents.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
