Mita Bhattacharya
21 October 2026
Simplify your enterprise kubernetes and govern your AI agents with Canonical
Discover how Canonical is delivering simpler kubernetes to scale enterprise AI at Kubecon North America (NA) 2026
Enterprises have a clear mandate: scale AI initiatives securely and without delay. Yet if the AI infrastructure is too complex, initiatives can be delayed, eroding business value. Kubernetes is often one of the key causes: 62% of organizations report that Kubernetes complexity is actively slowing or delaying business initiatives.
Canonical Kubernetes offers organizations a path to simpler infrastructure, combining upstream conformity, simplified lifecycle management, and long-term security maintenance. It helps enterprises reduce operational overhead and run workloads consistently across public clouds, private infrastructure, and the edge. At KubeCon NA 2026, Canonical will be showcasing how simpler Kubernetes can deliver supported, low-touch operations.
Get a preview of the demos we’ll be running at Kubecon NA in this article.
Governing AI at scale
Running hundreds or thousands of AI agents across an enterprise requires numerous policies and guardrails. Without safety measures, rogue agents may escape if deployed on the same infrastructure as the core production workloads.
To help organizations run autonomous AI agent fleets more securely, Canonical recently announced the alpha release of Charmed Openshell. At KubeCon, you can watch how Charmed Openshell works in action. By combining our low-touch, lightweight cloud platform MicroCloud and Canonical Kubernetes with Charmed Openshell, we preview an integrated approach to hosting and governing AI agents, equipped with the exact guardrails and policy enforcement needed to run autonomous agents in sandboxes at scale. If your organization is looking for a dedicated AI infrastructure cloud, make sure to check out our demo.
Achieving AI portability
AI applications are frequently locked into a single cloud large language model (LLM) provider. This means that switching models or migrating them to on-premises infrastructure – be it for cost, privacy, or compliance – requires extensive code rewrites. Engineering teams can reduce this vendor lock-in by implementing a private, portable, model-agnostic AI strategy using 12-factor application architecture. We’ll demonstrate how in this demo.
MicroCloud provides private cloud infrastructure on demand, while inference snaps run open source LLMs locally behind a standardized, OpenAI-compatible API. Canonical Kubernetes provides MicroCloud with supported, low-touch deployment, while Juju automates lifecycle operations for the entire stack. This architecture decouples model logic from application code, enabling LLMs to be treated as swappable configuration bindings. As a result, fewer application changes are needed when switching between compatible models and inference providers.
Simplifying end to end Machine Learning Operations (MLOps)
Setting up the comprehensive infrastructure required to train and serve custom AI models is notoriously complex. Organizations are often locked into expensive public cloud APIs just to manage datasets and training jobs. In this demo, we reveal a resilient alternative: Canonical Kubernetes equipped with Charmed AI components.
Rather than just walking through a standard deployment, we will show you what happens when enterprise models are pushed to the limit behind your firewall. You will discover how llm-d backed by vLLM intelligently separates prefill and decode phases to deliver true duplex performance with ease. Finally, we will put the self-healing power of Canonical Kubernetes to the ultimate test – you can watch how the system reacts in real time to a critical failure.
Securing the software supply chain
Cloud native engineering teams often struggle to balance developer velocity with container security. They might choose to settle for off-the-shelf container images that don’t quite meet their needs, use bloated images that have a large attack surface and create endless CVE noise, or opt for typical distroless images that may inadvertently introduce security blind spots. Canonical offers a better solution.
Watch live in action as we introduce rocks – hardened, Open Container Initiative (OCI) compliant images that go beyond distroless builds, combining the benefits of minimal container images with full auditability. We’ll show how to use our open source tooling to build your own rocks from the ground up, ensuring you only include the exact packages you need, while retaining the metadata required for accurate security scans. When your rocks are deployed onto Canonical Kubernetes, you establish a hardened, auditable environment from the build pipeline straight to the cluster.
Conclusion
For IT decision makers responsible for their company’s competitive advantage and budgets, the mandate is no longer about adopting Kubernetes – it is about controlling kubernetes at scale. Your teams should not have to choose between deploying generative AI or maintaining a trusted, governed environment. Canonical Kubernetes delivers simplicity through low-touch snap-based updates, automated Day 2 operations via Juju, and stability for 15 years with Long Term Supported (LTS) releases. The combination of MicroCloud and Canonical Kubernetes can provide organizations with a strong, stable low operations execution engine so your teams can focus on scaling AI.
Experience all our demos firsthand at KubeCon NA 2026 in Salt Lake City between 10-12 November. Stop by the Canonical booth to see our demos, or skip the queues and dive straight into your architectural requirements by scheduling a dedicated 1:1 strategy session with our cloud-native experts.
