As enterprises increasingly adopt cloud-native technologies, the benefits of AIOps and Agentic AI are becoming ever more apparent. These innovative technologies can provide intelligent analysis of Kubernetes cluster health, enabling organizations to autonomously diagnose platform problems and orchestrate issue resolutions with minimal human intervention. During the KubeCon + CloudNativeCon North America 2025 Conference, industry experts Vikram Venkataraman from AWS and Srikanth Rajan from Salesforce shared their insights into how Salesforce is pioneering the development of self-healing systems using AIOps and AI Agents.
The robust AIOps architecture created at Salesforce supports the Hyperforce Kubernetes Platform, a managed Kubernetes solution that operates across multiple clouds—AWS, GCP, and Alicloud. This architecture boasts impressive operational scale, with approximately 1,400 Kubernetes clusters, millions of pods, and thousands of compute nodes. Utilizing over 40 operators and integrations and 200+ monitoring plugins, Salesforce anticipates a five-fold increase in capacity over the coming years. The overarching goal of their innovative solution is to allow application teams to concentrate on delivering business value rather than getting bogged down by infrastructure overhead.
Venkataraman and Rajan introduced a comprehensive approach to Kubernetes platform operations, leveraging generative AI and multi-agent collaboration for effective cluster management. Their solutions significantly reduce key metrics like Mean Time to Identify (MTTI) and Mean Time to Resolve (MTTR) for critical cluster issues. Central to their Agentic AI solution are AI Agents designed with specific objectives, enhancing the AIOps platform and utilizing telemetry data to perform key actions—ranging from rolling back upgrades to automating troubleshooting processes.
The speakers shared insights into the challenges they faced when building intelligent AI systems for operations. A significant hurdle was designing the communication protocols among different agents and establishing appropriate guardrails and security permissions. They provided a detailed overview of their solution architecture, hosted on the AWS cloud platform. This architecture incorporates an AIOps UI for engineers, a Collaborator Agent, Amazon Prometheus and its agent, Amazon EKS, the K8sgpt Operator focused on MTTI metrics, and the ArgoCD Controller.
Furthermore, the discussion delved into the various layers of their tech stack, combining advanced open-source technologies with proprietary tools. Highlights included:
- Substrate: Kubernetes cloud platforms such as Amazon EKS, self-managed Kubernetes, Google GKE, and Alicloud ACK.
- Standard Capabilities: Essential features like storage, networking, autoscaling, DNS, load balancing, mesh, and Ingress. Technologies in this layer encompass Istio, Cluster Autoscaler, CSI, OPA, Ingress, CNI, LBC, and CoreDNS.
- Custom Integrations: This layer includes capabilities for identity management, secrets management, implementing guardrails, and log collection.
- Platform Capabilities: Key components for platform abstraction, deployment orchestration, lifecycle automation, visibility and observability, resiliency, cost management, and best practices enforcement, utilizing tools such as Argo, Kyverno, Spinnaker, Helm, Kube Magic Mirror, Sloop, and Periscope.
- API Layer: Provides customer access services and houses the Control Plane, APIs, and self-service portals.
To address various operational challenges like siloed tools, static workflows, and limited feedback loops, the Salesforce team devised an AI agents-based infrastructure management solution. They began with a limited set of AI agents, such as the AIOps agent for on-call reporting, a Kubectl agent that integrates with team communication channels like Slack to translate natural language queries into kubectl commands, and a Live Site Analysis Agent that automates the weekly platform availability review process.
Venkataraman and Rajan also discussed the concept of progressive autonomy in the adoption of AI-agent-based solutions within organizations. Initially, they integrated human oversight into the loop to ensure the safety and reliability of issue resolutions. As confidence in the AI agents grew, they began to grant more autonomy to these solutions, allowing them to take more significant actions without human intervention.
The pair emphasized that Salesforce has merely scratched the surface of what AI can achieve in operational environments. Their roadmap for the AIOps program indicates plans to scale AI agents to automate up to 80% of manual tasks, develop a knowledge graph to unify various components of their systems, and employ AI for the detection and troubleshooting of complex performance issues.
For additional insights and details on other conference sessions, attendees can visit the conference website and explore the full program schedule.
Inspired by: Source
