Automation
Groundcover launches shared workspace for AI agents and engineers
Agent Investigations coordinates multiple AI agents and human responders during production incidents, testing hypotheses against live telemetry data.
Key takeaways
- Groundcover's Agent Investigations creates a shared workspace where multiple AI agents and engineers collaborate on production incidents instead of working in isolation.
- Two specialized agents structure the work: an Investigator builds and tests hypotheses against telemetry data, while a Moderator validates evidence before accepting conclusions.
- The workspace supports external agents from Anthropic and OpenAI through a Model Context Protocol server, allowing multi-agent coordination on the same investigation.
- Hypothesis testing and assessments integrate directly with Slack threads, letting engineers guide investigations from within incident conversations.
- The system maintains customer data in their own cloud accounts while agents query it, addressing data residency concerns.
Observability provider groundcover has introduced Agent Investigations, a collaborative environment where AI agents and human engineers work together to resolve production incidents, according to details first reported by SiliconANGLE.
The workspace addresses a growing coordination problem: as teams deploy more AI agents for incident response, findings become scattered across private chats, tickets, and Slack threads. When production systems fail, multiple engineers often direct their own agents at the same problem, with each working in isolation and conclusions remaining invisible to the broader team.
How the system works
Groundcover automatically initiates an investigation when a monitoring alert triggers, though engineers can also launch one manually through the interface or via Slack. The company deploys two specialized agents to structure the work.
The Investigator agent constructs a graph of questions and hypotheses, then tests each against groundcover's logs, metrics, traces, and Kubernetes data. Each finding includes supporting graphs as evidence.
A Moderator agent reviews every proposed conclusion before it can be accepted. This agent is the only one authorized to close a hypothesis and rejects any claim lacking adequate supporting data. The separation between hypothesis generation and evidence validation aims to prevent unfounded conclusions.
Each hypothesis concludes with a structured assessment that either supports or refutes it, giving teams tested answers rather than unstructured generated text.
Multi-agent coordination
The workspace extends access to external agents, including Anthropic's Claude Code and OpenAI's Codex, through groundcover's Model Context Protocol server. Connected agents can read from and contribute to the same investigation. Agents focused on business intelligence and product roadmapping can also participate.
Hypothesis status and assessments post directly to the Slack thread tied to the original alert. Replies in that thread feed back into the investigation, allowing engineers to guide the work from within the incident conversation.
"Every engineer now has an agent, but each of those agents investigates on its own and findings end up in a chat window," said co-founder and CEO Shahar Azulay, who launched groundcover in 2021 with Yechezkel Rabinovich. "Agent 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."
Why it matters
As organizations adopt AI agents for operational tasks, coordination becomes critical. Without a shared workspace, teams risk duplicating work, missing insights, and losing institutional knowledge about how incidents were resolved. The structured approach to hypothesis testing also addresses a fundamental challenge with AI-generated conclusions: distinguishing between plausible explanations and evidence-backed findings. By requiring validation before accepting any hypothesis, the system creates an audit trail that shows not just the root cause but also the reasoning path and eliminated possibilities.
Architecture and availability
The investigations run on full-fidelity telemetry collected by groundcover's sensor. Because groundcover uses a bring-your-own-cloud architecture, customer data remains in their own cloud accounts even while agents query it.
Senior product manager Omer Mayer noted that engineers using Agent Investigations can watch as each hypothesis is supported or ruled out in real time, rather than waiting without visibility into agent progress.
The Investigator's graph serves as the incident record, showing how the team reached the root cause and which hypotheses were eliminated. Linking the monitor, Slack thread, investigation, and conclusion eliminates the need to reconstruct incident history from separate tools.
Agent Investigations is available to groundcover customers with AI-enabled backends. The company will demonstrate the capability at AWS re:Invent in Las Vegas from November 30 to December 4. Groundcover has raised $160 million to date, including a $100 million Series C round led by One Peak in July.
Details were first reported by SiliconANGLE.