Ubuntu Plans Gradual AI Features Built Around Local Inference

Canonical has detailed plans for integrating AI into Ubuntu over the next year, emphasizing a gradual approach rather than a single major release. AI features will be introduced only when mature, with a focus on local inference, open-weight models, open source tools, and clear interfaces to external services as needed.

Jon Seager, Canonical’s vice president of engineering for Ubuntu, outlined the plan in a recent Ubuntu Community Hub post. Seager stated that while Canonical is expanding its use of AI tools, it is not setting targets based on token usage or the proportion of code generated by AI. The focus is on ensuring engineers understand when AI tools are appropriate and when they are not.

“The bottom line is that Canonical is ramping up its use of AI tools in a focused and principled manner that favours open weight models with license terms that feel most compatible with our values, combined with open source harnesses. AI features will be landing in Ubuntu throughout the next year as we feel that they’re of sufficient maturity and quality, with a bias toward local inference by default.”

In other words, Canonical does not position Ubuntu as an AI-first operating system. AI will be added only where it enhances existing functionality and remains consistent with Ubuntu’s standards for security, privacy, quality, and open source principles.

So, looks like Canonical’s approach to AI in Ubuntu is more measured than that of other enterprise Linux vendors. While companies like Red Hat and SUSE promote AI as a core platform feature, Canonical is taking a more cautious, OS-level approach, introducing AI only when it can be delivered responsibly.

Moreover, Canonical divides future AI work into implicit and explicit features. Implicit AI enhances existing operating system functions without altering the user experience, such as improved speech-to-text, text-to-speech, accessibility, and screen-reading. Seager characterizes these primarily as accessibility improvements rather than AI-branded features.

At the same time, the explicit AI features will be more visible to users and may include agentic workflows, such as document creation, application development, automated troubleshooting, personal automation, or system administration assistance. Canonical notes that robust security and confinement controls are necessary before these features are widely implemented.

Local inference is central to Canonical’s plan. The company highlights “inference snaps” as a method for providing local access to models optimized for specific hardware. Users can install an inference snap to receive optimized components for their device, avoiding manual model downloads and configuration. These snaps are governed by Snap confinement rules, restricting model access within the system.

Another very interesting point that the post discusses is user control. In response to a question about disabling AI entirely, Seager stated that Ubuntu will not run models in the background “for the sake of it.” However, he does not expect Canonical to implement a global AI kill switch, as this would be difficult to enforce given the varied ways Ubuntu users interact with software.