Groundcover, an observability company based in Tel Aviv, has raised $100 million in a Series C round led by One Peak, signaling a significant shift in how observability tools are built and deployed. The round brings the company's total funding to $160 million. Morgan Stanley Expansion Capital, Zeev Ventures, Angular Ventures, Heavybit, and Jibe also participated. Groundcover was founded in 2021 and announced a $35 million Series B round in April 2025.

The new capital is intended to help Groundcover meet growing demand from engineering teams that need to observe more than conventional application performance. As agentic AI systems become a regular part of the software development lifecycle, companies are looking for visibility into what those systems actually do in production. That includes knowing which agentic workflows run, which models they call, which vendors they use, and what kinds of data those interactions generate. Groundcover's platform was built on open-source eBPF and OpenTelemetry technologies, and that foundation is now helping the company respond to the rise of AI-centric operations.

Observability moves beyond post-production monitoring

For much of the past decade, observability was associated with post-production work. Teams deployed an application, monitored it in production, and used traces and logs to find root causes when something went wrong. The goal was typically to reduce the time it took to resolve incidents and to keep services within acceptable performance thresholds. That approach is increasingly being complemented by a new kind of observability, one that pulls production context into earlier stages of development.

The rise of agentic AI is reshaping that workflow. Instead of monitoring a predictable request path from a user through a service and into a database, engineers are now watching AI agents use tools, call models, interact with other software, and make decisions. The process is less deterministic than traditional request flows. It generates larger volumes of telemetry and often touches sensitive data in ways that traditional dashboards were not designed to handle.

Groundcover CEO and co-founder Shahar Azulay described the current moment in observability as fascinating because it is no longer just about tracking latency and error rates. Teams are also tracking token usage, model behavior, hallucination rates, and the movement of data into and out of AI-powered systems. In his view, this is not simply a new kind of application performance monitoring. AI observability will become its own discipline because the underlying activity is fundamentally different.

Why eBPF matters for AI workloads

The technical foundation of Groundcover is eBPF, or extended Berkeley Packet Filter. That Linux kernel technology allows code to run safely inside the kernel without requiring a custom kernel module. It was originally used primarily for network monitoring, but Groundcover uses it to observe application and infrastructure activity across the stack. The important benefit is that eBPF works below the application layer, so it does not depend on a developer instrumenting each service by hand with an SDK.

According to Azulay, that design removes a major burden from engineering teams. With eBPF, there is no need to ask developers to change their code base or add an agent to every service. The observability layer can still see what is happening across an entire environment because it is embedded at a lower level. This approach was useful when teams had to monitor microservices and containers. It is even more useful now that AI tools are being adopted quickly and engineering organizations may not have full knowledge of every AI-powered workload running in their own environments.

Azulay compared the current visibility gap to the problems teams faced a decade ago before observability tools matured. Modern engineering environments are becoming harder to track because agents can spin up workflows, call external services, and use new models without waiting for a human to define a monitoring path. eBPF acts as a safety net in those cases. Even if a workflow was not manually instrumented, ground-level kernel data can still reveal which agentic workflows are running, which models are being used, and which vendors are involved.

The limits of distributed tracing for agentic workflows

Distributed tracing has long been a core part of observability. It follows a request as it moves across services so engineers can see where time is spent and where a failure occurs. Traditional tracing assumes a reasonably predictable number of hops. A request might go from an API gateway to an authentication service, then to a cache, and then to a database. That structure makes it possible to map a path and find the root cause of a slowdown or failure.

Azulay argued that this assumption breaks down once agents are involved. A single agent session may generate a very large number of tool calls, internal model calls, and decision points. The paths are not fixed. One session may take one route, while another session with the same starting prompt may go down a completely different path. That makes conventional traces much harder to interpret and much less useful as a way to understand system behavior.

In addition, the data inside an agentic trace may be different from a traditional trace. A trace might include the customer's actual prompt, which is sensitive content rather than merely a structured metadata field. This creates new privacy challenges because observability pipelines must be able to handle that data without exposing it to unauthorized parties. Groundcover says its architecture is designed to address this problem by storing telemetry inside the customer's own cloud environment instead of a shared vendor backend. That allows engineering teams to keep larger and more sensitive telemetry volumes within their own perimeter.

Azulay said the expectation now is that organizations will need to save more telemetry and also save it more privately. Agentic workloads generate richer operational data, and much of that data has security or compliance implications. Moving all of that information to an external observability platform can be impractical. By keeping the telemetry local, Groundcover is positioned to support AI observability without forcing companies to make a choice between visibility and data protection.

Agent Mode and MCP integration

Groundcover is not only changing its platform for AI workloads. It is also using AI to improve how engineers interact with observability data. The company has built an assistant called Agent Mode that lets engineers ask questions about their systems, build dashboards, and troubleshoot issues in logs and traces without manually writing queries. This is designed to help less experienced users get value from observability quickly and to help experienced engineers move faster during incidents.

Groundcover has also built an integration with the Model Context Protocol, or MCP, which allows AI agents to exchange context with external tools. The MCP integration connects Agent Mode to coding agents and workflow tools such as Linear. This makes it possible for engineers and AI systems to share information during an incident in a more natural way. Instead of forcing users to move back and forth between an observability dashboard and an AI-powered coding tool, the context can flow between them.

Azulay said adoption of the MCP integration has been faster than the company expected. Customers are using it in different ways depending on how far they are along in their AI adoption journey. Some use it to ask questions about system behavior without opening the Groundcover dashboard. Others use it to identify a problem and then move directly to writing a fix. This pattern points to a broader industry trend in which developers are beginning to create autonomous software development structures, with AI agents taking on more responsibility for both analysis and remediation.

Groundcover at a glance

• Founded: 2021
• Total funding: $160 million
• Latest round: $100 million Series C, led by One Peak
• Other investors: Morgan Stanley Expansion Capital, Zeev Ventures, Angular Ventures, Heavybit, Jibe
• Headquarters: Tel Aviv, Israel
• CEO: Shahar Azulay
• What they do: Observability technology built on eBPF and OpenTelemetry

The funding announcement comes at a time when the wider observability market is being reshaped by the shift from static production monitoring to dynamic AI-enabled development. Traditional application performance monitoring tools are still necessary, but they are becoming only one part of a larger picture. Engineering teams now need to understand the behavior of AI systems that plan, reason, and act on their own. That requires observability platforms to see activity that was not instrumented in advance, protect the data generated by AI interactions, and provide ways for both humans and machines to act on that information.

Groundcover's approach combines eBPF-based data collection with a customer-local architecture and AI-driven assistants. The company is betting that the combination will make it a central player in the next phase of observability. That phase is defined less by monitoring and more by the need to understand, secure, and manage AI infrastructure as it becomes embedded in every part of the software lifecycle.

Source: Network World News