Speak at The Fifth Elephant 2026 Annual Conference

Abstract

Modern AI agents can process logs, metrics, dashboards, and incident tickets, but often struggle with real-world operational reasoning due to limited understanding of underlying infrastructure.

Virtual networking layers abstract away much of the system complexity, while observability data alone does not expose topology, dependencies, or traffic flows needed for full context.

This talk explores how harness engineering can bridge this gap by combining topology, observability, and operational tooling to build infrastructure-aware agents that can explain failures, understand dependencies, and support incident response and modernization efforts across hybrid and multi-cloud environments. It also shares lessons learned from building such systems in Nutanix Flow and the challenges that remain in practice.

What We Will Discuss

• Using Infrastructure Topology as a Reasoning Layer
• Understanding Infrastructure Relationships
• Incident Investigation and Remediation
• Migration and Modernization

Audience

This talk is intended for:

• Early-career engineers interested in networking and infrastructure, anyone who wants to understand how modern virtual networking works, even without deep networking expertise.
• Network Engineers, SREs, Platform Engineers, Cloud and Infrastructure Architects, AI Engineers building operational copilots and agentic systems.

Key Takeaways

• Importance of infrastructure context in agentic systems
• Graph-based modeling of topology and dependencies
• Harness engineering patterns for connecting agents to infra tools
• Visualizing and explaining network behavior across environments
• Translating infrastructure across VMware, AWS, GCP, and Nutanix
• Real-world challenges in deploying AI-assisted operational systems

Why This Matters

As enterprises adopt agentic systems for automation and incident response, these systems often lack awareness of the underlying infrastructure that produces the signals they analyze. By combining topology, dependency mapping, and operational tooling, we can build infrastructure-aware agents that are more explainable, trustworthy, and effective—moving beyond telemetry-driven automation to systems that understand the full context of infrastructure behavior.

About the Speaker

Mansi Sharma is a Member of Technical Staff in the Nutanix Flow R&D team, specializing in virtualization and complex distributed systems. Her current focus lies in designing and implementing advanced networking features, including multi-cluster Kubernetes networking and VM Live Migration (LMCT).

Passionate about bridging the gap between observability and operational intelligence, Mansi is focused on building infrastructure-aware AI systems that help engineering teams better understand and navigate complex hybrid and multi-cloud environments.