Guest: Bassam Tabbara
Project | Company: Crossplane | Upbound
Show: KubeStruck
Topic: Kubernetes, Platform Engineering

Crossplane just achieved CNCF graduation status, joining an elite group of projects deemed production-ready for enterprise adoption. For Co-Creator Bassam Tabbara, this milestone represents more than a vanity badge—it signals that the cloud infrastructure industry has converged around a new approach to managing multi-cloud environments. In a conversation at KubeCon, Tabbara (CEO and Founder of Upbound), explained how Crossplane is evolving from a simple infrastructure control plane into the foundation for platform engineering and AI-driven operations.

From Container Orchestration to Infrastructure Control

Crossplane emerged from a simple observation: Kubernetes revolutionized container orchestration by introducing declarative, reconciliation-based management. You tell Kubernetes what you want running, and it autonomously manages those workloads. Tabbara and his team asked a fundamental question—why limit this powerful model to containers?

“When we started Crossplane, the idea was to bring that model, but instead of just limited to container workloads, to actually apply it to cloud and infra,” Tabbara explained. The result is a Kubernetes-native control plane that manages AWS, GCP, Azure, and other cloud infrastructure using the same declarative approach that made Kubernetes successful.

This wasn’t a simple engineering exercise. Tabbara acknowledged that managing diverse cloud infrastructure required building an entire ecosystem. “There’s a lot of infra to manage. You have to get an ecosystem playing on this, and it takes a long time to get a set of vendors to agree on how to build providers and functions,” he said. The team also had to modify Kubernetes itself to handle heavy CRD workloads, contributing changes back to the project and factoring Crossplane providers into smaller, more manageable units.

The Platform Engineering Surprise

While Crossplane began as an infrastructure management tool, it evolved into something more significant. Organizations discovered they could use Crossplane compositions to create their own API abstractions—internal platforms that expose Kubernetes-style APIs to developers without exposing underlying infrastructure complexity.

“We built this thing called compositions, which enabled not only managing infrastructure yourself, but creating your own compositions that are also exposed as Kubernetes APIs,” Tabbara said. “To our surprise, that really took off.”

This unexpected adoption positioned Crossplane as an early enabler of platform engineering, a movement that has since expanded to include tools like Backstage and Argo. Organizations are building intermediate platforms on Crossplane, creating self-service infrastructure that developers can consume through familiar Git-based workflows.

Why AI Needs Crossplane

The emergence of AI and agentic AI has created new requirements that align perfectly with Crossplane’s design. Tabbara outlined three ways the project intersects with AI adoption:

First, organizations are using Crossplane to manage AI infrastructure itself—provisioning GPU clusters for inference and training, scheduling workloads across those clusters, and handling the complex lifecycle of AI workloads. “If you wanted to bring up a set of clusters that you want to do inference on, or train models on, that’s something that Crossplane does really well,” Tabbara noted.

Second, and perhaps more importantly, AI agents require strong APIs to be effective beyond simple code generation. “AI agents work really well when they talk to really strong APIs,” Tabbara emphasized. “Most organizations that are seeing success with AI and AI agents today probably stop at code generation. They’re not at the point where it’s doing operations and deployment and diagnosis of things.”

Without platform APIs, agents can’t complete the full software development lifecycle—they remain stuck in the early stages. Crossplane’s API-first architecture provides the interface that agents need to move from code generation to actual deployment and operations.

Third, Crossplane is exploring running LLMs inside control loops themselves. “LLMs can actually do really interesting things inside the control plane,” Tabbara said. “Diagnosis, looking at log metrics, suggesting incidents—all of that work can happen as part within the control plane itself.”

What Graduation Means

For Tabbara, this is his second graduated CNCF project, and he understands the significance beyond the ceremonial aspects. “It’s a signal to the rest of the ecosystem that this is a strong project with vendor-neutral governance, that production deployments are continuous and growing in large organizations and small organizations alike,” he explained.

Graduation doesn’t mark an ending—it validates a beginning. Tabbara sees opportunities to build domain-specific control planes on top of Crossplane’s generic foundation. “There are interesting verticals that can be built on top of Crossplane for specific domains,” he said.

The project also aims to expand its role in AI operations, ensuring that agents can handle not just code generation but deployment, operations, and SRE workloads. The community continues to grow, with new contributors joining weekly.

Advice for CNCF Projects

When asked what advice he’d give to projects seeking graduation, Tabbara emphasized fundamentals: “Focus on building community, focus on getting maintainers, focus on running a clean open source game. Be clear about your governance structure upfront—your governance.md file that lives in the repo is as important as any other file there.”

But community alone isn’t enough. “Get into real customer use cases early. That’s the thing that keeps everyone honest,” Tabbara said. “It’s really easy to get engineers that get excited about code, but it’s way stronger when the code is actually useful to certain companies.”

The convergence of cloud infrastructure, platform engineering, and AI represents a rare moment when disruptive technology aligns with existing foundations rather than replacing them. As Tabbara put it: “Even with something as disruptive as AI, what we’re seeing is that it’s converging on Kubernetes and cloud native. It’s able to withstand something as large as AI—which is a great sign.”