What's happening with Ubuntu and AI (and why it's a problem)

This week, Canonical's vice president of engineering Jon Seager posted on Ubuntu's Discourse an overview (and later clarification) of what Canonical envisions with regard to generative AI tools in and around the Ubuntu Linux distribution. It's long and broad, so I'm breaking it down and sharing my thoughts topic-by-topic.

AI is coming to Ubuntu

In a testament to the fact that AI has become a marketing term of murky meaning, Seager's post touches on all kinds of topics, from code contribution guidelines, to theoretical accessibility enhancements, to job descriptions for Canonical employees, to end-user hardware limitations.

TL;DR: Canonical employees will be using LLMs in their development of Ubuntu, and LLM-based features for post-Resolute Racoon Ubuntu Desktop and Server are coming. None of the details are clear yet.

LLM-generated code contributions

Seager explained Canonical's view of LLM use in the development of Ubuntu. It's encouraging of LLM-assisted work, but in classic corporate fashion, definite expectations and guidelines aren't clear.

Canonical doesn't want its employees token-maxing, but it does want employees to integrate it into their workflows. In what way and to what extent is up to employees to discover.

Seager acknowledges the risks of slop and cognitive surrender. His solutions are careful review of everything LLMs produce and education on critical thinking for contributors. If and how these solutions will be implemented remains, again, unclear.

• The curl project has faced uphill battles recently with AI-generated bug reports.

Essentially: You can expect Ubuntu's code to be increasingly informed by LLM outputs in the coming year.

My take: It's weird how he emphasizes caution and thoughtfulness while also listing "accelerate development tasks" as a benefit of AI. Also, I don't like how Seager pits workers against each other by rehashing a line we've been hearing from the executive class since the AI rush started in 2022:

AI is not going to take software engineering jobs at Canonical, but other software engineers who are highly competent with AI tools certainly could.

AI features coming to Ubuntu

Seager throws out ideas for future "context-aware" Ubuntu Desktop and Server iterations with integrations that can automate common tasks. He differentiates between implicit and explicit AI features: some "implicit" features will use generative and predictive AI on the backend while "explicit" features will be explicitly, usually "agentic," AI tools.

Essentially: Expect features debuting with Ubuntu 26.10 similar to what you've seen on Windows.

Implicit features

As examples of implicit features, Seager mentions text-to-speech and speech-to-text accessibility tools.

My take: There are already multiple open source, Linux-native tools that do these things. I've written about SpeechNote, to name on example. If Canonical wants to clone them into the Ubuntu OS, though, that's a fine choice.

Explicit features

Seager lists explicit AI usecase examples we've been hearing about for years:

... authoring new documents or applications, automating troubleshooting workflows or even personal automation tasks such as targeted daily news briefings.

Later, he gets more Linux-specific:

Imagine being able to ask your Linux machine to troubleshoot a Wi-Fi connection issue, or to stand up an open source software forge that’s pre-configured, secured, and reachable over TLS. One could easily imagine using such a capability as a gateway for controlling your Linux machine from other devices through a variety of mediums - be that a mobile app, text messaging, voice commands or otherwise.

My take: I wonder if Seager realizes that shell scripts can already automate the configuration of remote servers. Take DietPi's software scripting as an example. Plenty of mobile apps let you remotely control your server, too.

• Also, I'm yet to see proof LLMs are trustworthy in managing server security.

Types of models Canonical will use

Seager indicates that while connections to cloud-based models will be manually configurable, Canonical will prioritize local, offline models. They'll be open weight models with licensing and harnesses that are friendly to open source values, packaged as Snaps.

He correctly acknowledges that models with open weights are not the same thing as open source models.

• "Open weight models" are ones whose parameters are accessible to anyone but whose training material and other important elements remain closed.
• "Harnesses" refers to interfaces built around the model that limit the model's environmental and tooling access.

Essentially: Expect more proprietary code in the Ubuntu ecosystem.

Context added: Canonical has a history of inviting the ire of FOSS purists. Its signature software package management system, Snap, uses closed backend servers.

My take: Why is Canonical leaning on open-washed models instead of developing one of the first high quality, truly open source models? That would make more of a splash in the AI world than yet another Copilot clone.

How you can opt out

In the later reply, Seager indicated that AI features will essentially always be an opt-in feature. This is in part because it'd be too cumbersome to ship Ubuntu with huge local models attached.

If you try LLM tools and later regret your choice, your "kill switch" will be simply removing the Snaps they came packaged in.

My take: That's at least easier than avoiding Copilot on Windows, I'll give them that.

Plans for consumer hardware

Seager acknowledges that current high quality local models require substantial computing power not everyone has access to. His solution is waiting for smaller models to gain better efficiency and for more efficient silicon to arrive, and collaborating with Ubuntu's silicon partners to reduce this wait time.

Essentially: Don't expect to be able to use many of Ubuntu's AI features on your old laptop any time soon.

Context added: The kinds of PCs that Ubuntu and other Linux distros have famously enabled the continued use of are PCs that are outdated and underpowered. Also, PC hardware costs have famously skyrocketed in recent months thanks to increasing AI usage.

My take: This is all so out of touch. Waiting has so far done the opposite of enabling me or anyone I know to make new and more powerful computer purchases. Personal computing as a lifestyle, in fact, feels under threat because of data centers buying up that silicon.

The bottom line

Enterprise customers that can actually afford AI-capable hardware may be pleased to hear all of this. Catering to them is Canonical's prerogative, of course.

• As an individual PC owner, though, I sense Canonical becoming increasingly as context-unaware as Microsoft.

Don't take my word for it. Read Seager's Discourse post and his later clarifying reply, then draw your own conclusions about Canonical's direction and Ubuntu's future.