AI in observability in 2026: Huge potential, lingering concerns

The role of AI in observability is evolving rapidly, but the data from our fourth annual Observability Survey makes one thing abundantly clear: the potential is real, and so are the reservations.

Practitioners overwhelmingly see value in using AI to help surface anomalies, forecast and spot trends, assist with root cause analysis, and get new users up to speed quicker. Of course, aspirations are not the same as outcomes, and the data—collected as part of the largest community driven survey on the state of observability—also points to concerns about letting AI act autonomously and without oversight.

In this blog post, we'll dive deeper into what the wider observability community thinks about AI and how it fits into their workflows and their future.  Based on data collected from the 2026 Observability Survey brought to you by Grafana Labs, we'll look at sentiments about potential use cases, trust, blockers, and the adoption of observability to monitor AI workloads.

Note: Check out the full 2026 Observability Survey to see what more than 1,300 respondents have to say about AI, open source, the expanding scope of observability, and the practices organizations are adopting today. You can also read our in-depth analysis on the state of open source in observability, or analyze the stats yourself in our Grafana dashboard, which includes filters for industry, role, region, and company size so your team compares to its peers.

The AI data and the times we live in

Before we dig into the data, let's first state the obvious: AI is evolving quickly—very, very quickly. 

Since we collected survey responses from last October to January, AI models have improved dramatically, and there has already been a seismic shift in agentic workflows going from concept to reality for many teams. 

That said, the Observability Survey was designed to understand what people want from AI, regardless of what tool or models they deploy: What types of AI capabilities would be valuable for observability? What guardrails do you need to establish trust? What would prevent you from using it? Luckily those questions—and the data that we collected from them—hold up even as the technology evolves.

AI could provide a welcome productivity boost to observability

We asked about six potential use cases for incorporating AI into observability platforms, and the overwhelming majority respondents think it would be valuable¹ in each scenario:

• Generate dashboards, alerts, or queries: 92%
• Surface anomalies and other issues before they cause downtime: 92%
• Forecast and spot trends: 91%
• Assist with root cause and correlation analysis: 91%
• Help new users quickly understand the relevant parts of the system: 89%
• Take autonomous actions (i.e., automatic remediations or triggering workflows): 77%

And while there's broad consensus, some use cases are seen as more valuable than others. For example, 75% of respondents say the ability to surface issues before they cause downtime is either "very valuable" or "critical to their use case," compared to just 63% who say the same about helping new users quickly understand the relevant parts of the system.

The trust gap: AI assistance vs. AI action

Skepticism about using AI for the use cases outlined above is extremely low (only 4% - 5%). The exception, however, is AI taking autonomous actions, where that number jumps to 15%. Autonomous actions also ranks last in perceived value, with just 49% saying it's either very valuable or critical. Organizations rely on observability to keep business-critical workloads up and running, so it makes sense that practitioners are still wary of offloading key decisions to AI.

Company size also plays into comfort levels with autonomous AI workloads. There's greater reticence to use AI for autonomous action at smaller companies, with 35% of respondents from companies with 100 or fewer employees saying they're either skeptical or see no value, compared to just 16% of respondents from companies with more than 1,000 employees. AI is often viewed as the great equalizer for small startups, so it's interesting to note the discrepancy between the two groups. 

The biggest AI blocker isn’t fear—it’s friction

More than a quarter of respondents (26%) say "too much manual input of required context" would be the most likely blocker to using AI for critical tasks—the most common response by a sizable margin. 

However, opinions change when you look at this question in relation to the use cases mentioned above. For example, if we go back to the most polarizing use case—AI taking autonomous action—those who say it would be critical to their use case cited manual input or context as the greatest potential blocker (32%). Those who see no value in AI taking autonomous actions are most likely to cite "AI breaking too much or not adapting" (31%) as the main blocker. 

This points to a divide between AI advocates and AI detractors, with the former more concerned with how AI fits within their organization and the latter more concerned about whether it actually works the way it's supposed to. 

AI isn't the only 'must have' in observability 

Only 15% of respondents say AI capabilities are an important criteria when selecting new observability tools, coming in as a lower priority than cost, ease of use, and interoperability. This a number that would presumably change if we asked people to retake the survey today, but it's still striking considering respondents could select up to three criteria.

There are some key differences of opinion based on roles. For example, directors of engineering are the most likely to cite AI as an important criteria (27%) while platform team members are the least likely to cite AI as a selection criteria (10%). 

Regardless, cost is universally cited as the top tool selection criteria. Undoubtedly as AI capabilities are leveraged more, they will also carry a cost. It will be interesting to see the interplay between these criteria in the months ahead, but it's safe to assume most organizations will continue to prioritize getting actual value from any AI services they adopt.

Accuracy and accountability are critical

You need to trust what you're seeing in order to make accurate and timely decisions about the operations of your systems. It's no surprise then that the old adage of "garbage in, garbage out" is carrying over to AI, with 95% of respondents saying it's important² for AI to show its reasoning. 

This includes showing sources, query logic, and confidence levels. It's a clear indication that practitioners want to maintain oversight over any actions AI takes, rather than just blindly trusting the results.

Those who see AI as critical to the previously discussed use cases are also the most likely to want AI to show its reasoning, while skeptics about those same use cases are the most likely to say it's critical that AI explains its reasoning. So whether you're on the fence or fully bought in, that need for transparency is high across the board.

How does observability of LLM-based applications fit in?

We asked about seven different emerging areas within observability, and observability of LLM-based applications was by far the least commonly adopted, with 29% saying it wasn't on their radar, and  14% using it for production workloads. Still, that's a big change from 2025, when 42% said it wasn't on their radar, and only 5% said they were using it for production workloads.

Admittedly, this is probably the least reliable statistic from this year's survey. Given the rapid rate of change, it's likely that many more organizations today are building agentic workloads and other LLM-based apps than they were just last quarter. For example, analyst firm Gartner, Inc., predicts 40% of enterprise applications will integrate with AI agents this year—up from less than 5% last year.

Or perhaps it's not that far off? Yes, we're seeing growing interest from our users, but the hype is always ahead of reality, and different organizations are at different stages of adoption in this evolving space. Regardless, this will serve a good marker to look back on to see just how much these numbers change in 2026.

Final thoughts on the Observability Survey and AI

We want to thank everyone who participated in this year's survey, which was our biggest yet. Between October and January, more than 1,300 observability practitioners and leaders from 76 different countries participated in our survey. We collected responses online and at events around the world, including our own shows and third-party conferences such as AWS re:Invent and KubeCON. 

It speaks to the power of the open source community that so many of you were willing to share your thoughts, and we take pride in making the Observability Survey an ungated, public asset for anyone who wants to learn from and share the data.

We're invested in making actually useful AI for our users, and listening to your feedback is  core to how we operate. What you share shapes how we work—and how we can do better. Thank you for sharing your knowledge and opinions with us.

^{1 "Somewhat valuable," "very valuable," and "critical for our use case" answers combined}

^{2 "Somewhat important," "important," "very important," and "critical" answers combined}

Grafana Cloud is the easiest way to get started with metrics, logs, traces, dashboards, and more. We have a generous forever-free tier and plans for every use case. Sign up for free now!