As enterprise platform teams move past generative AI experimentation and into production, the networking and security demands on Kubernetes are changing fast. F5’s 2026 State of Application Strategy Report found that 66% of organizations now permit AI to automatically adjust policies and configurations, putting new pressure on platform and DevOps teams to support AI workloads without slowing development or creating security blind spots. Ahead of KubeCon + CloudNativeCon North America 2026 in Salt Lake City, VMblog spoke with Buu Lam, Director of Community Evangelism, Open Technology & Market Strategy at F5, about how the company is meeting that moment, from open source NGINX to BIG-IP Next for Kubernetes running on NVIDIA BlueField DPUs.
In this pre-show Q&A, Lam explains why F5 sits directly in the data and inference path, and how token-aware routing, Gateway API Inference Extensions, AI Guardrails, and Model Context Protocol (MCP) inspection help govern a new class of agentic workloads. He also previews what attendees can expect at Booth #100 in the Salt Palace Convention Center, including a live demo of NGINX catching MCP tool failures disguised as HTTP 200 responses, whiteboard architecture clinics, and lightning talks, as well as F5’s sessions in the new AI Inference + Agentic track.
++
VMblog: What’s your elevator pitch for KubeCon + CloudNativeCon North America 2026? If attendees only remember one thing about your company after visiting your booth, what should it be?
Buu Lam: F5 covers Kubernetes end to end, from open source to enterprise scale. F5 NGINX runs inside the cluster, where platform and DevOps teams already work. F5 BIG-IP Container Ingress Services secures traffic entering the cluster, where NetOps and SecOps teams set policy. And F5 BIG-IP Next for Kubernetes accelerates and secures traffic for AI infrastructure on NVIDIA BlueField DPUs.
The takeaway? You can start with open source NGINX, grow into F5 NGINX Enterprise, and scale without having to re-platform as Kubernetes becomes business-critical.
VMblog: This is KubeCon’s return to Salt Lake City — what makes the North American cloud-native community particularly strategic for your business objectives this year?
Lam: Returning to Salt Lake City lines up with a massive industry turning point: enterprise platform teams are moving past generative AI experimentation and deploying AI models and autonomous agents into production.
In our F5 2026 State of Application Strategy Report, we found out that 66% of organizations now permit AI to automatically adjust policies and configurations. As autonomous workflows become more common, platform and DevOps teams face growing pressure to support AI workloads without slowing development or creating security blind spots.
KubeCon gives us an opportunity to hear directly from the practitioners navigating these challenges and understand how their experiences compare with what we’re seeing in our research.
VMblog: CNCF has introduced a brand-new “AI Inference + Agentic” track for 2026, covering GPU scheduling, model serving, and observability for production AI systems. Where does your company fit into that conversation, and what’s driving your investment there?
Lam: Silicon vendors and accelerator platforms provide the compute power behind AI, but platform engineers own the messy operational reality of getting traffic to and from models efficiently and securely. That is where F5 fits: directly in the data and inference path.
We’ve invested in F5 NGINX Gateway Fabric and Gateway API Inference Extensions to support token-aware model routing and load balancing. We’re also integrating F5 AI Guardrails and Model Context Protocol (MCP) inspection directly into the data plane to help teams secure prompts, detect prompt injection attempts, and monitor agent activity.
On the agentic side, open source NGINX now routes MCP tool calls to individual backends, applies per-tool rate limits and circuit breakers, and traces calls with OpenTelemetry. We’ll be demonstrating these capabilities at the show.
For AI factories and neoclouds, F5 BIG-IP Next for Kubernetes runs on NVIDIA BlueField DPUs, offloading networking and security from the host so more resources can be dedicated to AI workloads.
As AI inference moves into production, organizations need the same performance, reliability, and security they expect from their other business-critical applications. We’re bringing decades of experience solving those challenges to AI infrastructure in a Kubernetes-native way.
VMblog: Where exactly can attendees find you on the show floor at the Salt Palace Convention Center, and what’s the one must-see demo or experience you’ve built that showcases your technology’s real-world impact?
Lam: You can find us at Booth #100 on the main show floor at the Salt Palace Convention Center.
Our must-see demo tackles one of the newest operational headaches in Kubernetes: governing autonomous agent workflows and Model Context Protocol (MCP) traffic. Attendees will see open source NGINX inspecting MCP tool calls in real time.
The demo shows how NGINX catches tool failures disguised as HTTP 200 JSON-RPC responses, enforces granular per-tool circuit breakers, and pairs with F5 AI Guardrails to help prevent prompt injection, credential abuse, and unintended data modification before an autonomous agent executes an unvetted command against internal databases.
VMblog: Walk us through your booth strategy — what interactive elements, live demonstrations, or hands-on experiences will help attendees truly understand your solution’s differentiation?
Lam: We designed Booth #100 around live demonstrations and opportunities for attendees to connect directly with our engineers:
- Interactive Demo Pods: Hands-on testbeds covering real-world Gateway API migration paths, LLM latency optimization, semantic prompt caching, and multi-cluster ingress policy automation.
- 1:1 Whiteboard Architecture Clinics: Platform teams and architects can sit down directly with F5 principal engineers to talk through their specific Kubernetes bottlenecks, multi-tenancy constraints, or GPU ingress topology challenges.
- Technical Lightning Talks: Short, practical 15-minute briefings delivered continuously throughout the show floor hours by F5 engineering leads and community contributors.
VMblog: Are you hosting any booth talks, mini-sessions, or expert Q&A opportunities? What knowledge can attendees gain without attending a formal session?
Lam: Yes. Our Booth #100 theater features a curated schedule of practical sessions running across Tuesday, Wednesday, and Thursday.
We’re hosting technical sessions alongside key ecosystem partners, including Red Hat, Nutanix, vCluster, alongside NGINX open source project maintainers. Attendees can drop by without a formal reservation to learn about:
- Token-aware routing patterns and dynamic inference load balancing.
- Automating the migration from legacy ingress configurations using ingress2gateway.
- Applying policy-as-code to multi-tenant Kubernetes clusters running AI workloads.
- Exporting end-to-end OpenTelemetry traces from agentic tool calls directly into Prometheus and Grafana.
VMblog: The industry narrative has shifted from “Kubernetes for web scale” to “Kubernetes for AI scale.” How has your technology evolved to meet that shift, particularly as inference is projected to overtake training as the dominant AI compute workload?
Lam: Web scale was about stateless request/response balancing, predictable payload sizes, and horizontal pod autoscaling based on CPU and memory thresholds. AI scale introduces completely different dynamics: stateful prompt contexts, long-lived streaming token responses, and more complex demands on GPU infrastructure.
F5 evolved by introducing token-aware traffic management across BIG-IP Next for Kubernetes and F5 NGINX Gateway Fabric. Instead of routing traffic blindly based on round-robin algorithms, our platform inspects streaming connections, evaluates real-time GPU cluster utilization, enforces tenant token consumption quotas, and implements semantic prompt caching.
This helps reduce redundant model computations and improve inference efficiency.
VMblog: With NVIDIA’s growing role in the CNCF ecosystem (including its Dynamic Resource Allocation driver donation to Kubernetes SIG Node), how — if at all — does your company work with or compete alongside major silicon/accelerator vendors in this space?
Lam: Our relationship with the accelerator ecosystem is deeply collaborative. Silicon providers build the high-performance hardware and runtime environments that execute AI models; F5 delivers the high-availability networking, security, and traffic steering required to feed those systems reliably.
NVIDIA is a featured partner at our Booth #100 theater, and our work together includes running F5 BIG-IP Next for Kubernetes runs on NVIDIA BlueField DPUs to offload networking and security from the host. We also work with Red Hat on OpenShift, Nutanix on the Nutanix Kubernetes Platform, and with vCluster on multi-tenant AI infrastructure.
These partnerships bring together the compute, networking, and security capabilities organizations need to run AI workloads at scale.
VMblog: Agentic AI workflows are emerging as a new class of Kubernetes workload, with autonomy and orchestration requirements that differ from traditional services. How is your platform adapting to support these agentic systems?
Lam: Autonomous agents introduce a brand-new infrastructure challenge: the risk of “borrowed authority.” Employees grant coding agents and autonomous workflows access to enterprise credentials, enabling those agents to make tool calls across internal APIs, external SaaS platforms, and Model Context Protocol (MCP) servers.
F5 addresses this by sitting inline within the network interaction path. Using NGINX’s low-latency JSON parsing and native MCP header inspection, we can reclassify agent tool calls before they’re executed.
We apply least-privilege policy enforcement, validate user context and OAuth tokens, enforce rate limits, and sever connections if an agent attempts an unauthorized file read, privilege escalation, or destructive database write.
VMblog: Is your team presenting any sessions, participating in panels, or leading workshops — including within the new AI Inference + Agentic track? What knowledge will attendees gain from your speaking engagements?
Lam: Yes.
- Breakout Session:“Who’s Governing the GPU Lane? Policy as Code for AI Inference Traffic in Kubernetes”
- When: Tuesday, Nov. 10, 2026: 11:25 a.m. – 11:50 a.m. MT
- Where: Cosmic Canyon, Level 1, Halls 1–5, Salt Palace Convention Center
- What Attendees Will Learn: How platform teams can use InferencePool and InferenceObjective custom resources within the Gateway API framework to enforce tenant isolation, manage traffic prioritization, and avoid silent policy bypass during GPU saturation.
- Demo Theater Session:“HTTP 200, Tool Failed: Taming MCP Agent Traffic with NGINX”
- Where: Show Floor Demo Theater
- What Attendees Will Learn: A live, technical walkthrough demonstrating how open source NGINX inspects Model Context Protocol (MCP) tool calls, surfaces masked errors, implements circuit breaking, and streams OpenTelemetry metrics into Prometheus and Grafana dashboards.
