Observability spent most of the past decade as a post-production discipline, catching outages and shortening the time engineers need to find a root cause. It largely answered what broke and why after code was already running. That focus is now moving backward in the development cycle as agentic AI systems enter coding, testing, and deployment, pulling production context into the tools engineers use every day.
This shift is helping to fuel demand for Groundcover, which this week announced a $100 million Series C round. The company builds its observability platform on open-source eBPF and OpenTelemetry technologies, and investors are betting that the next wave of AI infrastructure will need a new kind of monitoring.
Groundcover at a glance
- Headline: Groundcover raises $100M as observability pivots from monitoring to AI infrastructure
- 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
Funding and market context
The new financing arrives roughly 15 months after Groundcover raised a $35 million Series B round in April 2025. With the Series C led by One Peak, the company now reports $160 million in total funding. Morgan Stanley Expansion Capital, Zeev Ventures, Angular Ventures, Heavybit, and Jibe are also named as investors. The capital injection signals a broader market shift: observability vendors are no longer just selling dashboards for IT operations teams; they are becoming part of the infrastructure that lets autonomous software systems run safely.
Groundcover CEO and co-founder Shahar Azulay says the industry is in the middle of a meaningful transition. “I think what is happening to observability right now is fascinating,” Azulay said. As AI agents become more capable, understanding system behavior requires more than simple service-level metrics. Engineers need visibility into model calls, tool selections, token consumption, response quality, and the flow of data between AI components and traditional systems.
How eBPF works and why it matters now
Groundcover is built on eBPF, short for extended Berkeley Packet Filter. The technology has long been central to Linux network monitoring, but it has broader uses. eBPF allows code to run safely inside the Linux kernel without requiring a custom kernel module. That means observability data can be collected at a very low level, underneath the application, without imposing much overhead on the systems being watched.
The critical benefit for Groundcover is that teams do not need to instrument each service by hand. Many traditional observability tools require developers to install an SDK, change code, and redeploy applications before monitoring can begin. Groundcover removes that step. “You didn’t have to have the developer instrument an SDK, change their code base, and so on,” Azulay explained.
This approach is becoming even more relevant as engineering organizations rush to adopt AI-powered development tools. Teams often move quickly enough that they lose track of what is actually running in their own environments. Azulay compared that gap to the visibility problems teams dealt with a decade ago, before modern observability tooling matured. eBPF operates below the application layer rather than relying on code written by a developer, so Groundcover can observe workflows that nobody thought to instrument in advance.
“eBPF is kind of that security net of even if you didn’t instrument, even if you’re not in full control, you’re gonna know which agentic workflows are running in production, which models are using, which vendors they’re using, and so on,” Azulay said.
From monitoring to AI infrastructure
The shift in observability is not just about adding a few AI-specific charts to an existing dashboard. Agentic AI systems behave differently from conventional software. A single agent session can spin up a large number of tool calls, internal model calls, and decision points with no fixed execution path. Traditional distributed tracing assumes a request will move through a predictable number of hops, such as an API calling a cache and then a database. Engineers can follow that path because it is stable enough to map by hand. Azulay says that assumption breaks down when agents are involved.
“With LLMs and agentic workflows, this is becoming very complicated,” Azulay said. An agent might call a language model, then decide to query a database, then call an external API, then revise its own plan, then call another model, and finally produce a result. The trace of a single activity can be massive, branching, and different every time. Existing tracing tools were not designed to make sense of that level of complexity.
The observability industry is responding by expanding what it measures. Teams now track token usage, hallucination rates, and model choice alongside traditional latency and error rate metrics. They also need to know which vendors are being called, how much those calls cost, and whether an AI system is consuming data that it should not have accessed.
Azulay does not think this is just an extension of application performance monitoring. “It’s not going to be the same product,” he said. “AI observability is not exactly APM.” The new discipline must handle unstructured decisions, large volumes of telemetry, and sensitive data embedded inside traces.
Privacy, data placement, and customer-side storage
One of the biggest challenges in AI observability is privacy. A trace that was once made up of structured request data may now include an actual customer prompt. That prompt can reveal business logic, personal information, or proprietary data. Shipping it to a third-party observability backend creates a serious compliance risk.
Groundcover stores telemetry inside the customer’s own cloud environment rather than in a shared vendor backend. This architecture supports the larger telemetry volumes that agentic workloads generate without forcing companies to send sensitive data off-site. It also makes it possible for AI agents to query observability data from the same environment where they operate.
“I think people are expected to save more telemetry, and save more telemetry more privately,” Azulay said. Keeping data close to the workload is especially important for regulated industries and for teams running models that use internal or customer data.
Agent Mode and the rise of Model Context Protocol
Groundcover is not only adapting its platform for AI workloads; it is also using AI to improve the experience of engineers. Agent Mode is the company’s built-in AI assistant. Instead of forcing engineers to learn a query language, Agent Mode lets them ask questions about system health, build dashboards, and troubleshoot problems across logs and traces. It can act as a conversational layer over the entire observability dataset.
The company has also built a Model Context Protocol integration, commonly called MCP. MCP is becoming a standard way for AI applications to connect with external tools and data sources. Groundcover’s integration connects Agent Mode to coding agents and workflow tools, including Linear. That allows engineers and AI systems to share context during an incident without switching between separate applications.
Azulay said adoption of the MCP integration has moved faster than the company expected. Customers are using it in different ways depending on how mature their AI workflows are. Some engineers ask questions through MCP instead of opening the Groundcover dashboard. Others use the integration to let an AI assistant investigate an alert and then write a fix directly. In both cases, observability stops being a passive tool that only humans consult and starts becoming part of an autonomous engineering workflow.
This pattern is not unique to Groundcover. Azulay described it as an industry trend, pointing to the broader move toward what he calls autonomous software development. When coding agents operate in production, they need continuous feedback from logs, metrics, and traces. Exposing that data through MCP gives agents the context needed to reason about system behavior and take action.
“People are basically starting to build their autonomous software development structure,” he said.
For observability vendors, the challenge is no longer just visualizing what happened after an incident. It is providing the deep, private, machine-readable context that AI systems need while also giving human engineers a simple way to ask for help and get useful answers. Groundcover’s latest funding round is a bet that the companies responsible for that next layer of AI infrastructure will look very different from traditional monitoring vendors.
Source:Network World News