How Agentic Observability Can Stop IT Teams From Drowning in Telemetry

Enterprise AI systems are increasing computing costs—a reflection of the high volume of telemetry data that teams must process and store. 

Every microservice, container, API call, model inference, database query, and network hop produces logs, metrics, traces, and events, which are then exponentially multiplied by the high volume of requests. This overflow of machine data also compounds the challenge that IT managers face in resolving incidents quickly amid growing architectural complexity.

Recently, ManageEngine, the enterprise IT division of Zoho Corporation, announced new causal intelligence and autonomous AI capabilities in Site24x7, its full-stack observability platform. The release addresses the complexity of hybrid cloud environments, where massive volumes of telemetry overwhelm human operators.

In an interaction with AIM, Srinivasa Raghavan, Director of Product Management at ManageEngine, notes that hybrid and cloud-native architectures have made IT operations highly interconnected, increasing the pressure on IT managers to resolve outages, downtime, and incidents.

That focus on interconnected systems and user impact is not new for Zoho. It stems from its own operational failures in the mid-2000s. Raghavan’s team developed Site24x7 after experiencing firsthand the limitations of reactive monitoring inside Zoho’s production environment. At the time, Zoho’s flagship CRM product would fail without the team realising until customers complained.

The company did have internal monitoring tools in place, but those systems primarily tracked infrastructure health rather than real user experience. The tools confirmed that servers were running, yet customers still experienced latency. That gap—between system health and customer perception—proved costly.

Raghavan explains that this disconnect directly impacts business outcomes. “When there is sluggish performance in an application, people will switch to different providers. You are actually losing your customers now. So any innovation, any application, or any service—the user experience matters.”

That experience shaped Site24x7’s evolution from infrastructure monitoring to full-stack observability anchored in business outcomes. 

The new update introduces domain-aware causal correlation that groups anomalies across applications, infrastructure, and networks into a single context-rich incident, reducing fragmented alert streams into an actionable problem statement. It deploys predictive anomaly detection alongside service dependency mapping to surface not just abnormal behaviour but also its upstream cause and downstream impact.

The platform also adds customisable AI agents that operate within defined guardrails, using approved solution documents to guide investigation and response steps. An MCP (Model Context Protocol)-enabled agentic layer standardises how agents access observability data, enforce governance controls, and execute tasks with auditability built in.

Site24x7’s agentic workflows attempt to codify how experienced engineers investigate incidents. The system queries APIs, evaluates transaction traces, inspects infrastructure metrics, and isolates database or code-level bottlenecks in sequence.

“It can actually make queries to our service, include that part of the analysis, and provide recommendations—what is the issue? Is it a server-level issue or a code-level issue that is impacting it? It will go step by step, mimicking how a DevOps or SRE engineer would analyse the problem to identify the root cause,” Raghavan says. The goal is not to replace engineers, but to reduce cognitive overload and compress investigation time.

While ManageEngine’s strategy begins with user-experience anchoring and governed automation, other vendors, such as data platform Splunk, have approached the AI-era observability problem from a different angle.

Splunk’s MachineGPT—a foundational model for machine language—focuses on teaching AI to interpret machine data as a structured language. Rather than centring business impact first, it concentrates on understanding the grammar and semantics of logs, metrics, and telemetry at scale.

Mangesh Pimpalkhare, Senior VP and General Manager at Splunk, explains the challenge and the need for such a model. “Large enterprises are operating with an ever-expanding digital footprint. They’re trying to raise the bar on quality and speed while keeping everything resilient and operationally sound. The problem keeps growing.”

Security and IT teams often receive dozens of notifications for a single incident, many of which are false positives or derivative effects. MachineGPT can learn those patterns to distinguish noise from meaningful signals.

Pimpalkhare describes the current burden: “They look at an incredible number of noisy alerts. And then the human operators have to sift through them, saying, ‘I’ve gotten 57 alerts in my queue.’ They drill down to see what is a false problem and what might be a misleading alert. Then, maybe one out of 10 turns out to be a real problem.”

MachineGPT correlates these alerts automatically, collapsing redundant signals into coherent incidents and reducing manual triage.

The aim is to compress resolution cycles dramatically. As Pimpalkhare puts it, “Imagine now, with AI tools, you start getting to these in seconds and minutes.” Speed matters because operational teams currently spend disproportionate time on routine triage rather than systemic improvement or innovation.

However, both Splunk and ManageEngine recognise that autonomy must be calibrated. In high-risk domains such as security, unrestricted automated action is not acceptable.

“They [engineers] don’t necessarily want AI tools to take action on their own anytime soon in areas like security,” says Pimpalkhare. “There might be other use cases—for example, when a server is completely over-provisioned, and CPU utilisation is at 100%. Do you want someone to confirm, ‘Don’t throw more load at the server?’ No, you don’t. You can take that action on its own.”