An AI coding agent, working for eleven days under human direction, produced a working operating system kernel that can boot a full Linux userspace, according to a new report from CIQ. The project, called Ginux, is a research prototype. The report says it is not ready for production use and does not claim to match Linux.
What can Ginux run?
Ginux is an independent x86-64 kernel written mainly in Rust. It boots through UEFI, the firmware standard used by most modern computers, and implements its own core operating-system mechanisms. It also includes a Linux compatibility layer, which lets it run some software built for Linux without modification. In the experiment, Ginux executed selected unmodified Alpine Linux programs. It also booted a Rocky Linux 10 userspace running systemd, the service manager used by many Linux distributions, all the way to multi-user.target.
How was Ginux built with an AI coding agent?
The work was not a single prompt followed by a finished product. CIQ describes the process as Human-Directed Agentic Engineering, or HDAE. In this model, people define the architecture, the constraints, the milestones, the acceptance criteria and the points where work should stop. An autonomous loop then carries out implementation, runs the code, diagnoses failures and revises its own output. One persistent coding-agent task carried the project from start to finish over the eleven days.
CIQ says the method still depends heavily on people. The report argues that domain expertise and executable validation are needed to keep the agent's output aligned with what was intended. In other words, the agent can move quickly, but someone has to decide what counts as correct and check whether it is.
Is Ginux ready for production use?
The report is also careful about what Ginux can and cannot do. Its compatibility and hardware support are narrow. The kernel has not received an independent security audit. CIQ makes no claim that Ginux is ready for production or that it is equivalent to Linux. The company presents the project mainly as evidence of how quickly established systems mechanisms can be assembled with AI assistance, not as a replacement for existing operating systems.
The report is available as a PDF through CIQ's publication channels, where readers can review the full description of the experiment and its validation process.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.