TL;DR: AI can be available and responsive but not working as intended, a distinction that matters more as AI moves deeper into production. The state of observability in 2026 reveals a gap between AI ambition and execution across quality, agent behavior, security, cost, and business impact. Closing it requires observability that can verify AI behavior and its effects throughout the technology estate.
New Dynatrace State of Observability 2026 research exposes a tension in enterprise AI: 87% use AI agents for observability and IT operations, while 44% target fully autonomous operations within five years, compared with 12% today.
Yet respondents say humans verify an average 66% of AI-powered decisions, and only 27% currently address visibility into how agents decide and act. That shortfall reflects a broader pattern: across nine AI observability capabilities, execution trails perceived importance by an average of 14 percentage points.
For technology leaders, this execution gap is the area to watch: organizations are preparing AI agents to do more before many can clearly establish how those agents reached a decision.
Meeting these challenges requires two complementary capabilities:
- Observability of AI establishes evidence about AI behavior, quality, security, cost, and impact.
- AI-powered observability applies causal analysis, automation, and agents to improve operations and advance autonomy.
One helps determine whether AI is working as intended; the other helps people and agents operate the wider technology estate.
About the research: Dynatrace commissioned an online survey of 1,202 technology leaders at enterprises across 12 countries in August 2026. Unless otherwise stated, results reflect respondents’ reports about their organizations; percentages may not total 100% because of rounding or multiple-response questions. See the full report methodology for respondent criteria, participating markets, weighting, question wording, and applicable bases.
What the State of Observability 2026 reveals
| Executive concern | Research finding | What it means |
| Investment | 78% expect observability budgets to increase; among them, 96% say AI adoption is at least a moderate driver. | AI is changing what organizations require from observability. |
| Complexity | 43% cite AI, LLM, and agentic workload expansion as a leading source of technology complexity. | AI adds new demands to already complex technology estates. |
| AI behavior | 43% consider agent-decision visibility important; only 27% currently address it. | AI requires evidence that extends beyond conventional system health. |
| Autonomous operations | 44% target full autonomy within five years; 12% report reaching that level today. | Greater AI authority increases requirements for validation, integration, governance, and oversight. |
| Runtime security | 55% of security respondents experienced a zero-day incident in the previous year. | Security teams need runtime evidence to prioritize actual exposure. |
| Open standards | 77% have moved OpenTelemetry beyond evaluation; 20% have made it a strategic standard. | Open standards help maintain telemetry continuity across environments and lifecycle stages. |
How does AI change the definition of whether software is working?
AI workloads can meet conventional availability, latency, and error rates, but still produce inaccurate, unsafe, costly, or otherwise unintended results. Measuring the business impact of AI behavior illustrates the execution gap: 39% consider it important, while 24% currently address it.
AI performance depends on the applications, infrastructure, security systems, and services around it. Observability must connect that environment with AI behavior, resource consumption, and business impact to determine whether AI is working as intended.
Why is AI adding complexity before it helps address it?
AI introduces models, agents, interactions, and additional resource demands into already distributed environments:
- 81% report increased compute and infrastructure spending from AI workloads
- 77% report greater need for cost-allocation visibility
For executives, the implication is direct: teams need to evaluate AI performance, reliability, and economics in context with the applications and infrastructure that support it.
AI-powered observability can help teams apply causal analysis and automation to interpret that complexity and act on what matters. By showing what AI is doing, what it depends on, and what it costs, observability gives leaders a clearer basis for deciding which workloads justify their resource consumption and where costs need tighter control.
How does observability support greater AI decision authority?
As AI contributes more to the operational lifecycle, letting it take on more consequential decisions raises the need for validation, integration, and oversight. Leaders need enough visibility into AI actions and consequences to decide where autonomy can expand and where stronger controls remain appropriate.
- AI already contributes across the operational lifecycle: reported impacts span all nine measured areas, from predictive detection to compliance.
- Organizations use layered validation: respondents use nearly four methods to validate AI behavior on average, with accurate source-data validation the most widely used.
- Integration, not trust alone, constrains greater autonomy: integration complexity leads reported barriers at 31%, alongside skills, oversight, data quality, and explainability.
Observability of AI helps teams evaluate those actions and consequences; AI-powered observability applies causal intelligence and automation to improve operations and extend autonomy where appropriate.
How does observability help prioritize security risk in AI-era environments?
The same evidence principle applies to security: the quality of a risk decision depends on evidence that reflects what is actually happening in the running environment.
- 55% of security-qualified respondents experienced a zero-day incident in the past year.
- 45% say their organizations continuously monitor runtime environments for vulnerabilities that are actually exploitable or actively targeted.
Organizations are also extending security monitoring to AI workloads and agents while adding AI-powered detection.
Runtime context clarifies prioritization: static severity shows what could be vulnerable, whereas runtime evidence shows what’s exploitable, what’s affected, and where remediation matters most.
How do open standards help maintain continuous visibility into AI behavior from development through production?
Evidence about AI behavior also needs to survive changes in technologies, providers, and lifecycle stages. Open standards such as OpenTelemetry provide a vendor-neutral instrumentation foundation that helps preserve that continuity.
The research shows broad adoption, but less strategic standardization:
- 77% have moved OpenTelemetry beyond evaluation, but only 20% have made it a strategic standard
- 37% cite AI and LLM workload observability as a top adoption driver.
Those findings suggest OpenTelemetry is increasingly relevant to a comprehensive AI observability strategy, but instrumentation is only the foundation. Open standards can preserve visibility as AI moves across providers and from development into production; observability must still interpret that telemetry in operational and business context.
Observability establishes the fact basis for reliable AI decisions and measurable outcomes
Context-aware observability connects AI behavior with the systems, risks, costs, and outcomes around it. This awareness gives leaders a stronger basis for three decisions:
- Where AI is delivering enough value to expand
- Where greater authority is warranted
- Where risk or cost requires intervention
The opportunity is to make those decisions faster with reliable evidence of how AI behaves and what changes as a result.
FAQ
What is the difference between observability of AI and AI-powered observability?
Observability of AI continuously evaluates AI behavior, quality, security, cost, governance, and business impact to establish whether AI is working as intended.
AI-powered observability uses causal analysis, assistance, automation, and agents to understand technology environments, improve operations, and advance autonomy.
What is the gap between AI ambition and execution?
Across nine AI observability requirements, current execution trails perceived importance by an average of 14%. The gap spans capabilities including AI quality, agent decisions, security, cost, and business impact.
How does observability help make AI reliable?
Observability shows how AI behaves in the context of the systems around it. That helps teams verify results, investigate causes, identify operational consequences, and determine where AI needs intervention or can take greater responsibility.
How does observability help manage AI costs?
Observability connects AI workloads with resource consumption, supporting infrastructure, and operational impact. This helps leaders attribute costs, understand what is driving them, and decide which workloads justify greater investment or need tighter cost controls.
How does observability help prioritize security risk in AI-era environments?
Observability connects security signals with runtime evidence of actual exposure. This helps teams distinguish vulnerabilities that are merely present from those that are exploitable or actively targeted, understand affected services, and prioritize remediation according to immediate risk.
Why are open standards important for AI observability?
Open standards such as OpenTelemetry provide consistent instrumentation across providers, technologies, and lifecycle stages. This gives organizations greater flexibility to follow AI behavior from development and evaluation into production as their technology and tooling choices evolve.
Looking for answers?
Start a new discussion or ask for help in our Q&A forum.
Go to forum