Jon Seager (Jon Seager), Canonical’s vice president of engineering and technical leader of the Ubuntu project, summarized plans to integrate functionality based on large language models into the distribution. In the first phase, Ubuntu developers intend to use AI models to improve the existing functionality of the distribution, and then add separate AI capabilities and workflows to the distribution for users interested in using AI. AI capabilities will be integrated into the distribution gradually over the course of 2027, without coercion and as they are brought to readiness.
When choosing AI tools, preference will be given to open models distributed under licenses that meet the spirit of the distribution, as well as AI platforms based on open source. By default, AI models will run locally. Integration with external cloud AI systems will be in the form of a user-controlled option.
It is emphasized that the goal of the initiative is not to turn Ubuntu into an AI product or to promote AI for the sake of AI, but to selectively integrate AI capabilities into the distribution where it will truly benefit users. For those who do not wish to use AI, the ability to disable AI capabilities will be provided. Examples of the use of AI include processing voice commands, diagnosing network and system problems, setting up services, analyzing logs, performing routine tasks and auditing servers.
AI functionality added to the distribution is divided into implicit and explicit use of AI. In the first case, AI models expand existing functions without changing the way they interact with the user, for example, they are used for speech recognition and synthesis. In the second case, AI models are offered as independent solutions, for example, as AI agents for automating work, AI assistants for training, generating and analyzing content.
To ensure security and simplify installation, it was decided to deliver AI models along with tools for their implementation in the form of snap packages optimized for various hardware platforms. Models will run in an isolated environment that does not have direct access to the rest of the system, and AI agents will function in accordance with existing access control and audit mechanisms.
Within Canonical, it was decided not to initially rely on a specific AI stack, but over the next 6 months to provide teams with the opportunity to delve into the topic, try different AI stacks, choose the best AI solutions for their tasks, understand the strengths of AI tools and their restrictions. Company employees will continue to be judged on how well they perform their tasks, not on whether they use AI.