-

groundcover Launches Investigations, Bringing Evidence-Backed Root Cause Analysis to AI-Powered Incident Response

Investigator agents test hypotheses against groundcover telemetry, while moderator agents reject conclusions that lack evidence. Engineers and their agents collaborate on a shared investigation, with progress and findings synced to Slack in real time.

SAN FRANCISCO--(BUSINESS WIRE)--groundcover, the world’s leading bring-your-own-cloud (BYOC), eBPF and OpenTelemetry (OTel)-native observability platform, today announced Investigations, a shared workspace where groundcover’s agents, a team’s own coding agents such as Claude Code and Codex, and on-call engineers investigate production incidents together. When a monitor fires, groundcover automatically opens an investigation, tests possible hypotheses against full-fidelity production telemetry its eBPF sensor collects, and accepts a root cause only after a second agent confirms the supporting evidence.

Investigator agents test hypotheses against groundcover telemetry, while moderator agents reject conclusions that lack evidence. Engineers and their agents collaborate on a shared investigation, with progress and findings synced to Slack in real time.

Share

AI agents now take part in almost every stage of engineering work, including incident response. When production breaks down, several engineers often point their own agents at the same problem, and each works in isolation. Conclusions end up in private chats, tickets or Slack messages that aren’t visible to the entire team. And while agents can generate convincing explanations for production failures, a plausible explanation isn't necessarily the right one. Without shared evidence and a way to validate conclusions, teams are left reconciling competing answers rather than resolving the incident.

With Investigations, every step of the investigation is visible. Agents test hypotheses against production data, document what supports or refutes each one, and challenge conclusions that lack evidence. Hypothesis status and assessments post to the Slack thread tied to the alert, and replies in that thread flow back to the investigation. Engineers don't have to wait for agents to finish before evaluating their answers. They can follow the investigation as it unfolds, challenge findings, and redirect the work while it runs. External agents, including those used for business intelligence, coding, and product roadmapping, can connect through groundcover’s MCP server to read from and contribute to the same investigation. Every person and every agent on the incident works from one record of what was checked and what it showed.

“Every engineer now has an agent, but each of those agents investigates on its own and findings end up in a chat window,” said Shahar Azulay, CEO and co-founder of groundcover. “Investigations gives humans and agents one place to work an incident, where every hypothesis is tested against production data and every conclusion has to be backed by evidence. Teams don’t need to swap out their agents for ours, but they now have a shared source of truth. They can bring the agents they already use into a shared investigation where everyone can see what was tested, what was ruled out, and why.”

When Agents Multiply, Findings Become Scattered

groundcover built Investigations on the premise that every agent on the team, groundcover’s and the customer’s, should work from one set of evidence, and every conclusion should be proven before anyone acts on it. Rather than generating another incident summary, Investigations creates a structured record of the investigation itself, including the questions asked, hypotheses tested, evidence collected, and conclusions reached.

An investigation starts automatically when a monitor issue opens. Engineers can also launch one manually from the groundcover UI or from a button in Slack. From there:

  • Investigator agent: Builds a graph of questions and hypotheses, queries groundcover’s logs, metrics, traces and Kubernetes data to test each one and records findings with supporting graphs. Agents can examine application and infrastructure behavior together rather than relying on isolated signals.
  • Moderator agent: Reviews every proposed conclusion. Only the Moderator can close a hypothesis, and it sends back any claim without supporting findings. This separates generating a possible explanation from validating whether the evidence actually supports it.
  • Structured assessments: Each hypothesis ends in an assessment that supports or refutes it, so teams read a set of tested answers instead of pages of generated text. Engineers can see not just what the agents concluded, but the evidence behind each conclusion.
  • MCP access for outside agents: Claude Code, Codex and other agents connected through groundcover’s MCP server can read and add to the same investigation.
  • Two-way Slack sync: Hypothesis status and assessments post to the alert thread, and replies from the thread feed back into the investigation. Engineers can question findings and steer the investigation without leaving the incident conversation.

The investigation graph shows how the team reached the root cause, which hypotheses were ruled out, and what evidence settled each one. The monitor, Slack thread, investigation and conclusion stay linked, so nobody has to rebuild the record across separate tools after the incident closes.

Investigations draws on groundcover's full-fidelity logs, metrics, traces and Kubernetes context to test hypotheses against what actually happened in production. Rather than examining individual signals in isolation, agents can investigate behavior across applications and infrastructure, with supporting evidence attached to their assessments. Because of groundcover’s BYOC architecture, that data stays in the customer’s own cloud while agents query it.

“The most vulnerable point of an incident is often the initial wait, when agents are working and the on-call has nothing to look at,” said Omer Mayer, Senior Product Manager at groundcover. “With Investigations, engineers watch each hypothesis get supported or ruled out in real time. If the work heads the wrong way, they can step in and steer it. They can see what the agents are investigating, understand why they're reaching a conclusion, and step in before the investigation goes down the wrong path.”

Availability

Investigations is available to groundcover customers. groundcover will demonstrate Investigations at AWS re:Invent in Las Vegas, Nov. 30 - Dec. 4. Attendees are invited to stop by booth #1852 or schedule a meeting with our team to discuss their observability challenges.

To learn more about Investigations, visit https://www.groundcover.com/product/investigation and read our blog post.

About groundcover

groundcover is the full-stack observability platform for engineers & agents, allowing teams to monitor everything they run in their cloud, without compromising on data, control, or cost. It’s deployed inside the customer’s cloud premises and provides complete visibility into applications, infrastructure, and AI workloads without operational overhead. The platform offers unlimited data coverage at a fraction of the cost of legacy observability tools. Founded in 2021, the company is backed by leading global investors, including One Peak, Morgan Stanley Expansion Capital, Zeev Ventures, Angular Ventures, Heavybit and Jibe. Learn more at https://www.groundcover.com.

Contacts

Media Contact
Chris Churilo, VP of Marketing
chris.churilo@groundcover.com

groundcover


Release Versions

Contacts

Media Contact
Chris Churilo, VP of Marketing
chris.churilo@groundcover.com

More News From groundcover

You’re Invited: groundcover Takes Its Birthday Party to AWS re:Invent 2026

SAN FRANCISCO & LAS VEGAS--(BUSINESS WIRE)--groundcover, the world’s leading bring-your-own-cloud (BYOC), eBPF and OpenTelemetry (OTel)-native observability platform, is bringing its Birthday Party to AWS re:Invent 2026 in Las Vegas. Why a birthday? Because observability is being reborn, and groundcover plans to celebrate it with neon pink jackets, loud colors and the best vibe on the expo floor at booth 1852.The Occasion: Most booths at re:Invent will say the same five words, but groundcover’s...

groundcover Expands Its eBPF Sensor With Deeper Kubernetes and Infrastructure Visibility

SAN FRANCISCO & SALT LAKE CITY--(BUSINESS WIRE)--groundcover, the world’s leading bring-your-own-cloud (BYOC), eBPF and OpenTelemetry (OTel)-native observability platform, today announced a major expansion of its eBPF sensor, the component that runs in every Kubernetes cluster and collects telemetry without code changes. The sensor now ties application requests to the SQL queries they execute and adds support for ClickHouse and SIP protocols. It also stamps Kubernetes workload, namespace, cloud...

You’re Invited: groundcover Throws a Birthday Party at KubeCon + CloudNativeCon North America 2026

SAN FRANCISCO & SALT LAKE CITY--(BUSINESS WIRE)--groundcover, the world’s leading bring-your-own-cloud (BYOC), eBPF and OpenTelemetry (OTel)-native observability platform, is throwing a party at KubeCon + CloudNativeCon North America 2026. There is no birthday, and nobody will be forced to blow out candles in front of co-workers. groundcover simply thinks observability has earned a celebration, and it’s bringing demos, scratch cards and gifts to booth 111 to mark the occasion.The Occasion: See o...
Back to Newsroom