Enterprise & Work

87% of surveyed tech leaders report AI agent use, oversight lags

Dynatrace’s survey of 1,202 technology leaders finds widespread AI agent use in IT operations, but only 27% say they address visibility into agent decisions and actions.

Dynatrace Survey Finds AI Agent Adoption Outpacing Oversight
Dynatrace Survey Finds AI Agent Adoption Outpacing OversightLalmch / pixabay
By James Calloway5 min read

Why it matters

  • 87% of surveyed technology leaders reported moderate or extensive AI agent use for observability and IT operations.
  • Only 27% said their organizations address visibility into agent decisions and actions.
  • The August 2026 online survey covered 1,202 enterprise technology leaders across 12 countries.
  • People still verify an average of 66% of AI-powered decisions; 12% report fully autonomous operations and 44% target that level within five years.
  • 81% reported higher compute and infrastructure spending from AI workloads; 77% need better cost-allocation visibility.

Dynatrace found that 87% of surveyed technology leaders report moderate or extensive use of AI agents for observability and IT operations, while only 27% say their organizations address visibility into agent decisions and actions.

That 60-point spread is the report's central warning: adoption has run ahead of the records needed to judge what agents chose and whether their actions helped. For enterprises allowing software to recommend or execute operational changes, the gap matters because service metrics can confirm that a system responded without showing that an agent made the right call.

Dynatrace published the findings on November 10, 2026, in its State of Observability 2026 report. In that report, the company says respondents reported “moderate or extensive use of AI agents for observability and IT operations.” The narrower 27% figure covers whether organizations address visibility into agent decisions and actions.

How strong are the survey's findings?

The results come from an online survey Dynatrace commissioned in August 2026. It covered 1,202 enterprise technology leaders across 12 countries.

Respondents described their own organizations rather than behavior that researchers independently verified in production. Because Dynatrace sells observability software, the report is best read as a vendor-sponsored snapshot of the challenges its survey respondents describe, not a census of enterprise AI use.

The limitation narrows what the percentages can prove. They measure reported experience and stated plans, not deployed scale or measured reliability.

What does 87% adoption actually mean?

An AI agent is software that recommends or performs actions within a workflow. The survey's 87% figure covers observability (system monitoring) and IT operations. It does not mean 87% of companies have agents authorized to make consequential decisions on their own.

One organization might use an agent to summarize an incident and propose a response for an engineer to review. Another might allow an agent to carry out a predefined action inside a narrow part of its infrastructure. Both can report using agents despite sharply different levels of authority.

That distinction matters because more agents in a workflow can lift adoption figures without proving those agents can run critical work. A summary assistant and an agent allowed to alter production systems should not share the same readiness standard.

The headline number measures presence, not independence.

How autonomous are agents today?

Respondents estimated that people still verify an average of 66% of AI-powered decisions. Separately, 12% reported fully autonomous operations, while 44% said they were targeting that level within five years.

Dynatrace treats the five-year figure as a measure of ambition, not a forecast. It does not establish how many organizations will reach full autonomy by then.

Integration complexity was the most frequently cited barrier to autonomy, at 31%. Respondents also cited skills, oversight and data quality as challenges.

The 66% verification estimate reinforces the distance between deployment and autonomy. The survey does not say whether reviews were formal approvals, spot checks or close examination, so the figure measures the incidence of checking rather than its depth.

Current evidence shows continued human checking, while the five-year target remains intent rather than performance.

What can teams see after an agent acts?

Standard system monitoring can tell an engineer that a service responded and how long it took. Those signals do not reveal whether an agent selected an appropriate action or whether the action resolved the underlying incident.

If an agent recommends restarting a service, Dynatrace says a useful record would show which alerts informed the recommendation, whether a person approved it, and what happened afterward. Engineers could then check whether the restart helped or merely coincided with recovery.

That distinction turns observability into an audit trail (a record of actions and approvals) for automated decisions. Without the record, an apparently successful action remains difficult to evaluate against the task that prompted it.

Dynatrace found that 43% considered visibility into agent decisions an important observability requirement, compared with 27% that reported addressing it. The 16-point difference does not prove that the other respondents had no logs. It compares what respondents regarded as important with what they said their organizations were addressing.

The practical result is a priority gap, not proof that every organization lacks records. More respondents value decision-level visibility than report addressing it, leaving the control question unresolved.

Do agents actually reduce operating costs?

Some 81% reported higher compute and infrastructure spending from AI workloads. Another 77% said they needed better visibility into how those costs are allocated.

Dynatrace warns that measuring incident duration without capturing the extra resources an agent consumes can make automation look cheaper than it is. A tool that records both can show whether time saved offsets the added compute (processing capacity) and infrastructure cost.

Connecting those measures lets teams compare different approaches to the same work. The report does not quantify the resulting savings, so it supports a measurement method rather than a business-case estimate.

AI savings cannot be treated as proven when spending rises and cost attribution remains incomplete.

What must be true before an agent gets broader authority?

In a November 17 follow-up, Dynatrace outlined a progression from “gathering evidence to recommending actions, then acting with approval before moving toward broader autonomy where warranted.” The sequence places evidence before expanded permissions, rather than treating a successful pilot as proof of readiness everywhere.

For an operations team, Dynatrace says to measure an agent against a specific task. An agent that reliably categorizes alerts has not necessarily shown it can choose when to change a production system.

Broader authority requires three answers, according to the report: can the team reconstruct what the agent did, justify the authority it held, and measure whether its action helped? Each question targets a different weakness: unexplained behavior, excessive permission or an ineffective intervention.

Without those answers, higher adoption may reflect more agents in the workflow rather than evidence that they are ready to run it. That staged model makes the order explicit: establish evidence, test the task, then expand authority.

On Dynatrace's evidence, the next constraint on enterprise agents is not adoption alone; it is the ability to prove that an action was authorized, appropriate and effective.

Source: unite.ai

Share this article:

More from James Calloway

James Calloway

Show full bio

News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.

791 articles

Related articles

  1. Forrester names Dynatrace a Leader in AIOps Wave assessment
  2. BlackFog survey: 47% report sensitive data exposure via AI
  3. Only 10% of Enterprises Let AI Agents Decide Without Human Approval
  4. Coupa survey: 72% of leaders would let AI agents handle transactions
  5. PwC survey: AI accountability scattered, 11% say it's unclear

« Previous articleNext article »