Groundcover Raises $100M for AI Agent Observability

Someone Has to Watch What the Agents Are Doing

Observability company groundcover announced a $100 million Series C on July 29, 2026, led by One Peak with participation from Morgan Stanley Expansion Capital and existing investors Zeev Ventures, Angular Ventures, Heavybit, and Jibe. The round values the company at roughly $500 million — about four times its previous valuation — and brings total funding to $160 million. The pitch behind it is a question the industry has been quietly accumulating: once autonomous agents are doing real work in production, how does anyone actually know what happened?

• $100 million Series C led by One Peak at a ~$500 million valuation, roughly 4x the prior round
• Total funding now $160 million, following a year in which the company tripled annual recurring revenue and doubled headcount
• The platform now monitors AI agents in production, alongside conventional application and infrastructure telemetry
• Agent Mode lets engineers surface, troubleshoot, and remediate production issues through an agentic interface

Why AI Agents Break Traditional Observability

Classical monitoring assumes a fairly stable relationship between a request and the work it triggers. A user calls an endpoint, a handler runs, a database is queried, a response returns; traces capture that path and metrics summarize it. Agentic systems do not respect that assumption. A single instruction can fan out into dozens of model calls, tool invocations, retries, and sub-agent delegations, and the shape of that fan-out is decided at runtime by the model rather than by the code.

The result is a telemetry problem with two awkward properties. The volume is high and bursty, because one user action can produce a hundred spans. And the interesting failures are semantic rather than structural — the trace completed successfully, every span returned 200, and the agent still did the wrong thing. Monitoring built for the first kind of failure does not automatically catch the second.

What Is Agent Mode?

groundcover's answer includes Agent Mode, an agentic layer over the observability data itself. Rather than requiring an engineer to know which dashboard to open, it surfaces anomalies, walks the correlated telemetry, and proposes remediation. There is a pleasing symmetry to it — using agents to supervise agents — though the practical justification is more mundane. The data volume produced by agentic workloads is beyond what a human on call can reasonably scan during an incident.

The company also emphasizes privacy and control in its architecture, keeping telemetry inside the customer's own environment rather than shipping everything to a vendor's cloud. For teams running agents that touch customer records or internal systems, that is not a nice-to-have. It is often the condition of deploying at all.

The Growth Numbers Behind the Round

The financial context matters for reading the valuation. groundcover tripled ARR over the past year and doubled its global headcount, which is what turned a roughly $125 million valuation into a $500 million one. Observability has been a crowded, well-served category for a decade; growth at that rate generally signals a workload the incumbents were not built for.

What This Means for Teams Shipping Agents

The practical takeaway for anyone in our AI coverage audience who is moving agents from prototype to production: instrument before you ship, not after the first strange incident. The failure modes that matter — a tool called with subtly wrong arguments, a retry loop that silently burns budget, a sub-agent that returned plausible nonsense — are only visible if the traces captured enough semantic detail to reconstruct the decision.

That is the same discipline behind good agent sandbox design: assume the agent will do something you did not anticipate, and build the surfaces that let you see it when it does. A well-funded, purpose-built tooling layer arriving now is good timing for everyone building in this space.

Sources: BusinessWire — July 29, 2026; Calcalist CTech — July 29, 2026; TNW — July 29, 2026.