Rethinking AI Agent Telemetry: Groundcover’s Vision for Keeping It Within Your Cloud Environment

The landscape of AI agent observability is rapidly evolving, raising questions for businesses about the observability tools and solutions they truly need.

This week, the observability startup groundcover officially announced a successful funding round of $100 million led by One Peak, bringing its total capital raised to $160 million.

According to the company, it boasts over 250 paying customers and has significantly increased its annual recurring revenue in the past year. Furthermore, groundcover is becoming a preferred choice over well-established observability platforms within enterprise settings. Though these figures are based on company reports, they indicate significant growth in a highly competitive segment of enterprise software.

This market has traditionally been dominated by major players such as Datadog, Dynatrace, New Relic, Splunk, and Grafana, which collectively generate billions in annual revenue and have years of product development behind them. Entering this arena is no small feat.

groundcover posits that advancements in artificial intelligence have fundamentally altered the foundational principles on which these traditional platforms are built.

Rather than merely competing on features, the four-year-old startup aims to persuade businesses that the architecture of observability must adapt as AI systems become more autonomous, generate vast amounts of telemetry data, and increasingly engage in software operations. Whether this perspective is accurate remains uncertain, but it serves as an intriguing lens to examine the evolution of observability in conjunction with enterprise AI.

AI’s Impact on Telemetry Management

Historically, observability was seen as a post-deployment practice. Engineers would deploy applications, monitor logs, metrics, and traces, investigate incidents, and iteratively improve reliability.

However, this workflow is shifting.

With the advent of AI-assisted software development, deployment cycles have accelerated significantly. Coding tools are generating more code, infrastructure is evolving quickly, and organizations are deploying increasingly sophisticated distributed systems that integrate microservices, Kubernetes clusters, APIs, and large language models. Concurrently, businesses are beginning to utilize AI agents that perform complex workflows, interact with external tools, and engage with production systems.

Each of these activities generates considerable telemetry data.

This influx of operational data is driving organizations to retain it rather than discard it. The use of AI applications adds multiple layers of observability beyond traditional infrastructure monitoring—including prompt execution, model latency, token usage, retrieval pipelines, tool invocations, and agent behavior. As companies explore autonomous systems, this telemetry becomes invaluable for understanding an AI system’s actions and reasoning.

Consequently, this trend creates challenges with pricing models that charge based on data volume.

Historically, engineers have often responded by sampling traces, reducing retention times, or limiting which data gets collected. These practices help manage costs but also compromise visibility, especially when AI-driven systems require more comprehensive operational context.

“We’ve observed a significant surge in telemetry,” stated groundcover co-founder and CEO Shahar Azulay during a recent press briefing. “Users are dissatisfied with the value they’re receiving from platforms like Datadog, as they often find themselves limiting data collection.”

Whether such dissatisfaction will significantly influence the market remains to be seen, but the overarching trend is hard to ignore. AI is shifting the focus of observability from merely collecting sufficient data to ensuring organizations have access to all the telemetry they might eventually require.

Rethinking the Architecture of Observability

While several observability providers have introduced AI-driven assistants, automated root cause analysis features, and other AI functionalities in recent years, groundcover argues that these developments fail to address what it sees as a more fundamental concern: the location of telemetry and how customers pay for it.

Instead of relying on a conventional SaaS model that manages telemetry within vendor-owned infrastructure, groundcover adopts what it describes as a bring-your-own-cloud (BYOC) model. Customers retain control over the data plane, including telemetry storage and processing, within their AWS, Microsoft Azure, or Google Cloud environments. Meanwhile, groundcover offers a managed control plane and user interface, with an option for complete self-hosted deployments.

While some competitors, including Datadog, do present limited options for customer-controlled data residency, these do not equate to a full BYOC approach. In most scenarios, telemetry is handled within the vendor’s infrastructure, allowing only partial controls.

This design choice influences nearly every facet of the company’s strategy.

By allowing customers to use their existing cloud infrastructures, groundcover argues it can avoid pricing based on telemetry volume. Instead, its pricing structure primarily focuses on the number of monitored hosts, independent of telemetry volume.

This shift in pricing is said to influence customer behavior positively, as organizations can retain a complete dataset for operational review, compliance, and AI-based troubleshooting without worrying about cost prohibitions.

“Our pricing is based on infrastructure size, not data volume,” Azulay emphasized.

This distinction is significant since AI workloads can drastically increase telemetry flow compared to the infrastructure itself.

While this model may not automatically reduce costs for all users, companies with a lighter workload distributed across many hosts may experience different financial implications than denser Kubernetes environments. The company notes that host-based pricing is optimal for enterprises with high telemetry density and may be less favorable for less active systems.

Ultimately, the focus is not merely on costs but on predictability. Infrastructure teams within enterprises frequently wrestle with fluctuating observability costs in line with application growth. Groundcover’s model aims to correlate pricing with infrastructure planning rather than mere data volume.

Innovative Use of eBPF Technology

A second cornerstone of groundcover’s strategy is its utilization of eBPF, a Linux kernel technology that has gained prominence as a crucial component for modern cloud observability.

eBPF enables applications to monitor network traffic, system calls, and overall application behavior with minimal coding adjustments, negating the need for manual instrumentation.

This capability fosters quicker deployments and a wider coverage of telemetry across infrastructures.

For organizations leveraging Kubernetes and cloud-native applications, this reduction in instrumentation complexity can lead to faster deployment times and enhanced telemetry insight.

Azulay argues that this advantage becomes particularly vital as AI systems create increasingly intricate interactions across services.

“Our sensors allow in-depth monitoring from infrastructure to applications and AI tasks without the need for developers to modify their code,” he noted during the briefing.

Though eBPF itself is not unique—many observability vendors incorporate it—groundcover contends that its distinctive approach lies in the fusion of automated eBPF data collection, customer-controlled storage, OpenTelemetry compatibility, and unified pricing on a single platform.

The company’s briefing acknowledges that no single technology creates a competitive edge; rather, it is the integration of eBPF, BYOC infrastructure, host-based pricing, and comprehensive observability together that sets them apart.

Shifting to AI Agents as Active Participants

A particularly intriguing aspect of groundcover’s vision extends beyond traditional monitoring functions.

The company increasingly frames observability as infrastructure that supports autonomous software development.

While traditional observability tools serve human operators tasked with resolving production issues, groundcover believes future platforms will increasingly cater to AI agents.

Its Agent Mode product enables engineers to investigate incidents utilizing natural language queries across logs, metrics, traces, and Kubernetes events. More importantly, Azulay envisions observability evolving into a feedback mechanism that educates coding agents about the actual activities occurring in production environments.

Instead of simply identifying failures after deployment, observability can serve as a continuous source of operational context for AI systems to evaluate changes, spot regressions, and suggest or even implement fixes.

“We’re witnessing a transformation in observability, as it shifts from a post-deployment tool to a system providing production insights back to coding agents for enhanced code optimization,” Azulay commented.

For now, the company emphasizes that human involvement remains essential. Agent Mode investigates incidents and provides recommendations, but any changes to production still require human approval. Azulay anticipates a gradual increase in autonomy as organizations become more comfortable allowing AI systems greater involvement in operational processes.

This perspective illustrates a broader shift within enterprise software, where AI agents are increasingly integrated into development, testing, deployment, and operational processes, rather than merely acting as isolated assistants.

The Appeal of Alternatives for Enterprises

groundcover is venturing into a highly competitive market, filled with vendors who boast decades of experience in the enterprise sector.

In 2025, Datadog alone reported over $3 billion in annual revenue. Other major players like Dynatrace, Cisco’s Splunk division, Grafana Labs, and New Relic each maintain extensive partner networks, mature integrations, and robust enterprise support that newer entrants struggle to match.

groundcover does not aim to outpace these established companies overnight.

Instead, it posits that AI is instigating an architectural shift akin to past transitions from on-premises systems to cloud-native solutions.

Azulay notes that many customers initially turn to groundcover to cut observability expenses, but they often remain due to the desire for unrestricted access to enriched telemetry and AI-driven workflows.

He asserts that deployments usually replace existing platforms instead of existing alongside them, although no public data indicate customer migration rates or independent validation of these claims.

The company’s journalist briefing further advises scrutiny of any performance claims. Revenue growth, customer numbers, and enterprise adoption statistics stem from groundcover itself; customer case studies touting substantial savings are vendor-generated and should not be accepted as independent verification without additional evidence. The briefing also suggests vigilance regarding what metadata might leave customer environments in standard BYOC models, rather than assuming no data reaches vendor infrastructure.

These cautions are essential as the observability market grows crowded. Gartner is currently documenting over one hundred observability products, with nearly every major vendor promoting AI-powered capabilities.

Future success will likely hinge less on the question of AI’s importance—an increasingly evident reality—and more on whether enterprises determine that existing architectural frameworks remain adequate.

The Bigger Picture for Investors

Narrowly viewed, groundcover’s Series C funding round represents yet another significant round of investment in infrastructure.

However, from a broader perspective, it captures an ongoing debate about the future of observability in an era where software is increasingly self-writing, self-testing, and self-managing.

As AI continues to generate vast quantities of operational data, conventional views on telemetry collection, pricing, and storage may face mounting scrutiny. Vendors that rely on data ingestion revenues may need to adapt their economic models to align with shifting customer expectations. New entrants, on the other hand, have the chance to build their offerings based on these evolving assumptions from the outset.

groundcover believes this opportunity lies in weaving together customer-controlled environments, automatic telemetry gathering, and AI-optimized operations within a platform tailored for autonomous software rather than simply adding AI functions to legacy observability tools.

Whether this architectural gamble proves successful will depend on enterprise adoption trends in the coming years.

However, the company’s latest funding round suggests that at least some investors believe the next competitive landscape in observability won’t center solely on dashboards or alerts, but on crafting the operational data foundation that increasingly intelligent software needs to monitor and eventually oversee the systems it operates.