Virtana lets its AI agents pull in agents from other vendors to work IT incidents

Virtana has pushed its observability platform a step closer to automated incident response. The company added Agentic Observability Enterprise Management, which lets its AI agents bring in agents from other vendors when something goes wrong in IT.

That matters because enterprise AI agents have mostly been trapped inside one vendor’s box. If a failure cuts across several systems, each agent sees only a slice. The new setup is meant to change that: a Virtana agent that spots trouble can invite outside agents into the same investigation, as long as they support the Model Context Protocol or the open Agent2Agent protocol.

Virtana says that includes agents already bundled with software a customer owns. Once inside, they get access to the company’s monitoring data, which outside agents have been able to query since February through the System-Aware MCP Server. That feed includes more than 25,000 real-time metrics, and the agents also get the System Dependency Graph, which shows how applications depend on the infrastructure underneath them.

The company is aiming this at incident work that normally turns into a slow human war room. Chief Product Officer Amitkumar Rathi said that process can take hours to assemble and run, while a group of agents can do the same work in “milliseconds.” But the autonomy is not unlimited. It only applies to incident types a customer has already defined, and customers can keep engineers in the loop, require sign-off for each action, or both.

Virtana also built in control over who gets in and what they can do. Policy decides which outside agents are allowed into an investigation and how many tokens they can use once there. Every action goes into an audit log. In the company’s example, a third-party agent flags a slowdown, Virtana’s root cause analysis agent traces it to a database bottleneck after a software deployment, and the system confirms the application is back to normal after the fix.