AI enters production faster than IT leaders will let it fix incidents alone

Grafana Labs finds that organisations are rapidly deploying AI in observability, but confidence falls sharply when systems are given authority to remediate incidents without human approval.

Grafana Labs says the spread of artificial intelligence across observability is running ahead of confidence in letting those systems act without human approval. In its 2026 State of Observability survey, the company found that 85% of IT leaders are comfortable with AI carrying out read-only investigations, but only 52% are willing to let it remediate incidents autonomously. It said 68% of organisations have now moved AI beyond simple investigation and into pilots or production use.

That gap between diagnosis and action is central to the report’s findings. Anthony Woods, co-founder of Grafana Labs, said the real challenge is deciding how much authority to grant software once it begins operating inside production systems. He argued that there is a clear difference between using AI to explain what has gone wrong and allowing it to make changes on its own, and said future autonomous systems will need strong visibility into how applications and infrastructure behave before they can be trusted to act.

The survey also suggests that AI is already shaping day-to-day operations for engineering teams. Grafana Labs said 67% of respondents reported an improved on-call experience from AI-powered tools, while 28% said those tools had significantly reduced burden and alert fatigue. A separate discussion of the survey on OpsMatters said most respondents want AI to show its reasoning before acting, underscoring that explainability remains a major condition for wider trust.

Spending on observability is rising alongside that caution. Grafana Labs said 78% of respondents increased observability budgets over the past year, and 73% expect further rises in the next 12 months. The company added that AI is contributing to that spending both because teams are adding AI-specific features to observability stacks and because AI-assisted development is generating more infrastructure and application activity to monitor.

The survey also points to a broader platform shift. More than half of respondents said their organisations now use a unified observability platform across the business, and Grafana Labs said consolidation was rated very or extremely important by 79%. Among those that had already consolidated, success rates were higher than among organisations still split across teams. The study further found that 80% of organisations are already building or running products that use large language models, while many are extending monitoring to model quality, latency and token usage as AI systems move deeper into production environments.