CentOS founder assembles a Linux-compatible Rust kernel in 11 days with AI

October 27, 2026. Gregory Kurtzer, founder of CentOS and Rocky Linux, unveils Ginux, an x86-64 kernel written in Rust and designed to run Linux applications without relying on the Linux kernel. 11 days. The time it took an OpenAI Codex agent driving the GPT-5.6 Sol model to produce most of the code. 139,000 lines. The volume of retained code, spread across 1,037 commits at an API cost of about $8,000. Why it matters: an experienced developer with an agent has just prototyped a Linux-compatible kernel in under two weeks — a method that cuts straight to the future of systems development.

Human-Directed Agentic Engineering, the method

Kurtzer is not claiming the AI did the work alone. He describes the approach as Human-Directed Agentic Engineering (HDAE): a human defines the architecture, requirements, constraints, and acceptance criteria, while a coding agent handles implementation, testing, debugging, and revisions.

That nuance changes everything. The model does not “invent” a kernel: it implements a specification set by its human director. Executable tests act as the guardrail — every contribution the agent makes must pass them before being kept. The result is a working prototype, not a heap of loosely generated fragments.

What Ginux can already do

Ginux is not a proof of concept that prints a banner. It ships its own memory management, scheduler, process model, virtual filesystem, networking stack, device support, and a Linux ABI compatibility layer.

On the syscall side, it covers 375 calls from the Linux 6.12 x86-64 interface: 82 fully implemented, 245 partially implemented, and 48 deliberately denied. Concretely, Ginux runs Alpine Linux binaries built against musl and BusyBox, and Rocky Linux programs linked against glibc. It even boots a Rocky Linux 10 userspace through systemd’s multi-user.target stage.

The architecture is organized into domains, each isolating its own processes, address spaces, filesystems, mounts, capabilities, and lifecycle. Ginux can also pull and assemble OCI container images, using them to provide Linux userspace environments while it handles execution and isolation itself. Everything ships in a single GINUX.EFI image.

What is still missing

Kurtzer is the first to state the limit: Ginux is a research prototype, not a production-ready operating system. Linux ABI coverage is incomplete, hardware support is limited, and several areas — networking, filesystems, resource enforcement, security — still need substantial work.

Today, development and testing target QEMU/KVM virtual machines, with an x86-64 Q35 configuration, OVMF firmware, Intel IOMMU, and virtio devices. No GPU, no NVMe. The repository also documents emulated execution on Apple Silicon. This is a test bench, not a Debian replacement.

Why it matters for the Linux ecosystem

The Ginux experiment does not stand alone. In October 2026, Linus Torvalds was himself experimenting with AI-generated kernel code and measuring a 3x speedup. The question is no longer whether AI will write low-level code — it already does — but who directs it, who validates it, and who owns the result.

Ginux also joins a line of kernels written in Rust for memory safety, alongside projects like Asterinas. But its real novelty is not the language: it is the speed. Where a kernel historically took a team years, a seasoned developer with an agent produced a prototype booting systemd in 11 days.

Verdict

Ginux will not replace Linux, and Kurtzer does not claim otherwise. Its value lies elsewhere: it provides a public data point on what the HDAE method can produce in complex systems software, with checkable numbers — 139,000 lines, 1,037 commits, $8,000.

If you need a kernel to run in production, stay on Linux or a mature Rust kernel like Asterinas: Ginux is a research prototype. If you build critical systems, watch the method instead — a well-directed agent, bounded by executable tests and an experienced architect, can clear in days a volume of code that used to take months. The skill that is becoming scarce is not writing the code: it is knowing how to specify it, test it, and refuse it.

References

• Linuxiac — CentOS Founder Builds Linux-Compatible Rust Kernel with AI in Just 11 Days
• GitHub — gmkurtzer/ginux
• It’s FOSS — Rocky Linux Creator Built a New “Not Linux” Kernel in 11 Days Using AI
• TechSpot — CentOS co-founder creates a Linux-compatible kernel in just 11 days
• Zenodo — Ginux technical report (Gregory Kurtzer)