Canonical is attempting to solve the AI privacy paradox by wrapping local models in Snaps, prioritizing security at the potential cost of system performance. It is a pragmatic yet controversial bet on containerized intelligence that will likely further polarize the Linux community.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
Ubuntu is Putting AI Inside Snaps (Yes, Really)Added:
So, have you wondered what the future of AI in Ubuntu is?
Recently, Canonical released a blog post on what their plans are for 2026 and beyond of integrating AI into Ubuntu.
They've come out and said they're taking a big-picture approach, and they're going to choose open-weight models with licenses that line up with open-source values.
And what they're going to do is they're going to build on top of these open-source tools, and they are deliberately biased towards local inference, so your data stays on your machine by default. And the features will roll out when they're mature and solid, not rushed to market. For those that aren't aware, when someone refers to the fact that AI is a local inference model, it means that you can host the AI yourself on your own machine or on your own network, and you're not using something in the cloud.
How is Canonical using Internally, their focus is in education and real skill development, allegedly not vanity metrics, and their teams are encouraged to try different tools and find what works. And they say they found that AI tends to shine on those mechanical tasks when you find a free context because context is everything when it comes to AI. But if you force it everywhere, it doesn't help because there are some tasks it's better at than others.
And their performance reviews look at delivery quality, not how many tokens you burnt. And tokens is what the AI uses for input and output.
So, the bottom line is they're saying AI is not going to replace their engineers, but their engineers who master AI will have an edge over the others who do not.
They're convincing folks to probably deliver more with less, but who knows.
So, they've said that responsibility and transparency is the heart of the strategy. So, they don't want anything to do with random AI-generated pull requests. They know that folks relying on these tools stunt learning, and they want folks to continue being skeptical of AI.
And to them as well, the whole open-source is a complicated phrase when it comes to large language models because and that is this case, open weights matter on how the model was configured and tweaked.
And they're going to weigh the actual license terms, not just that the weights are downloadable because a model has its weights, it will it doesn't mean it's necessarily open-source.
So, they're saying that there is this big difference between implicit and explicit.
So, AI to quietly improve features that folks already use, speech-to-text, text-to-speech, smart screen reading, and easy accessibility when it's powered by local models. So, no new apps to learn, and things carry on as they do.
But the other big thing in AI and craze is the whole thing about agentic workflows, which is instead of having an AI sending and receiving text, it can do things on your behalf.
And there might be these features, of course, in future coming coming to Ubuntu, which authoring documents, building preconfigured servers.
But because of the risk of these, they would need to be strict security controls and opt-in and sandboxed cuz you don't want your AI to go delete your entire cluster of servers or delete all your documents.
So, they are building this with snaps.
It's a shocker, I know.
So, instead of folks having to use a llama or hanging they'll be able to do, for example, snap new atron 3 nano and get a model optimized for their machine.
And because they're snaps, they will have the same confinement rules as snap packages, so that means a model doesn't get free rein on your machine or your files.
Although, who knows.
And they are expecting more models in future.
So, as hardware gets better and strong partnerships going forward, they will try to implement stronger and better models.
And of course, their vision goes past all of this. Their long-term vision is Ubuntu machines to be queryable and assistive. So, in other words, instead of letting you get on with what you're doing, you could ask your machine to troubleshoot Wi-Fi in plain English, asking a server to analyze logs, and provide root causes to incidents.
Or instead of having to preconfigure Git, you could ask it to do that.
So, they're saying it would be read-only analysis and full audit trails, but the agent doesn't bypass security models it lives in.
So, very interesting.
But the other side, apart from the security side of it, and I won't even go into that whole debate about how intelligent AI models are, let alone the local models, is you need decent hardware.
Because if you're going to run a local model, you're going to have increased power draw. You're going to probably need a good graphics card because they tend to use those better.
And they're saying that they've got partnerships, so it will mean that Ubuntu will just be ready in the future and out the box, you get your brand new Ubuntu or Ubuntu machine with AI built into it, local only.
So, what does this mean for you? They're saying from 2026 onwards, they're going to ship AI features that are deliberate, secure, and aligned with open-source values.
Ubuntu should become more accessible and have easier accessibility in future, and they're saying that it won't be so-called slop, it'll be well-integrated AI.
How will you turn this off? Who knows.
It's a very interesting thing.
I must admit that words open-source and AI going together at the moment tend to go together as well as oil and water.
They tend to be extreme polar opposites of each other, not for everyone, but for a lot of folks, and I suppose the long-term thing is who's going to need an system administrator to look after your Ubuntu machine? Ubuntu would do it all itself from the glorious snap packages.
I'm starting to think it becoming a need to host snap packages might not be that far off. That's [snorts] another video coming in future.
So, what do you think about all of this?
Are you planning to use it? Are you not going to use it? Do you agree, disagree with the vision? Let me know in the comments. As always, thanks for watching and bye for now.
Related Videos
Agentforce NOW AMA: Build with React and Salesforce Multi-Framework
SalesforceDevs
490 viewsβ’2026-05-28
How agent o11y differs from traditional o11y β Phil Hetzel, Braintrust
aiDotEngineer
450 viewsβ’2026-05-28
Re: π£οΈπthepropheduπ2026 GST 103 CLASS (E-EXAM REVISION)
theprophedu
636 viewsβ’2026-06-04
WEB TECHNOLOGIES UNIT-2 | Degree 4th sem BCOM Computers web technologies unit-2 full explanationπ―β
LearnwithSahera
1K viewsβ’2026-05-29
More tests are always better? How to use AI to identify tests that bring little value
Alliance4Qualification
335 viewsβ’2026-05-29
Search Algorithms Explained in 60 Seconds! π€π¨
samarthtuliofficial
218 viewsβ’2026-06-01
People of Game of Thrones using JavaScript DOM
AltCampus
296 viewsβ’2026-05-30
Instagram accounts got PWNed
EricParker
13K viewsβ’2026-06-03











