1. 5:45 PM - 6:10 PM EDT

   Welcome & Opening Remarks – CNCF Toronto Team + Index Exchange

   in-person
   SPEAKERS
2. 6:30 PM - 6:45 PM EDT

   AI-powered OTel triage, right in your terminal

   in-person

   Setting up OpenTelemetry correctly is hard. Environment variables, SDK dependencies, collector config, endpoints: each with its own way to go wrong.

   In this demo, I'll show otel-checker and gcx (Grafana CLI) instrumentation check on a deliberately-broken OTel setup. You'll see it enumerate the findings, then turn them into a single prioritized plan covering every issue at once, either synthesized by Grafana Assistant (on a Cloud context) or aggregated locally from the explanation docs bundled with the checker (works offline, on OSS, on Enterprise).

   SPEAKERS
3. 6:45 PM - 7:00 PM EDT

   Secure, Stateful, and Serverless Agents on Kubernetes

   in-person

   Traditional Kubernetes pod-per-agent deployments fail at scale: running untrusted agents poses security risks, while idle agents waiting on LLM/MCP or user feedback waste massive cluster resources.

   In this talk, we introduce Agent Substrate, an open-source runtime that achieves high-density agent multiplexing (up to 30x oversubscription) on Kubernetes. Substrate combines gVisor/microVM kernel-level sandboxing with sub-second suspend and resume capabilities, preserving full volatile RAM and filesystem state across hibernation cycles without overloading the K8s API server.

   We will demo how Substrate rapidly hydrates and hibernates stateful agents, and show how Kagent layers on top, providing the control plane, abstractions, and UI required to orchestrate diverse harnesses (coding agents like Codex/Claude and framework agents like ADK/LangGraph) in a secure, serverless environment.

   SPEAKERS
4. 7:30 PM - 7:50 PM EDT

   Rethinking Identity, Trust and Security in Cloud-Native AI

   in-person

   As AI moves from isolated models to autonomous agents embedded within cloud-native environments, traditional approaches to identity and security are being challenged. AI agents can increasingly access APIs, invoke tools, interact with workloads and make decisions with limited human intervention. This session aim to explore how engineering teams can rethink identity, authorization, observability and trust for AI-native applications by addressing questions around agent identity, least privilege, workload interactions, runtime visibility and containment. The discussion will provide practical considerations for building secure, resilient cloud-native AI systems without slowing innovation.

   SPEAKERS
5. 7:50 PM - 8:05 PM EDT

   Dude, where did my tokens go?

   in-person

   Many organizations are rapidly building GenAI capabilities into their products, and thus providing better outcomes for their end-users. However, the operational cost of these new capabilities is not trivial. At the same time, many organizations have the need to implement safety checks on the GenAI queries of their users or the responses their users get. Therefore, we need robust observability of these new application capabilities.

   With the proliferation of frameworks, libraries and SDKs for building GenAI capabilities, it's very difficult to keep up with the instrumentation, such that we can build robust observability solutions. Moreover, developers building GenAI capabilities to their applications often have to use the programming language of their existing application code base. Therefore, even if we kept up with instrumentation for a certain programming language, for example Python, it is much more difficult to keep instrumentation up to date for all programming languages.

   I my talk I'll show how users struggling with these challenges can easily overcome them by using OpenTelemetry eBPF Instrumentation in their environment, to add instrumentation or augment their instrumentation for GenAI features. Since eBPF Instrumentation works at the protocol level, it doesn't matter what programming language is used or which programming framework is picked up by the development team.

   Attendees will learn:

   • How to instrument GenAI features in any environment with OpenTelemetry eBPF
   • How to capture and attribute GenAI operational costs and safety metrics
   • How protocol-level observability future-proofs your stack
   • How to deploy and validate GenAI observability in production
   SPEAKERS

No sessions match the selected labels.