Canonical's approach to AI is refreshingly thoughtful - Microsoft should take note

Elyse Betters Picaro / ZDNET

Follow ZDNET: Add america arsenic a preferred source connected Google.

ZDNET's cardinal takeaways

• Canonical lets you prime and take however you'll usage AI.
• In Ubuntu, AI is built into cardinal features and optional AI tools.
• While Microsoft is astir control, Canonical puts you successful charge.

In a caller blog post, Jon Seager, Canonical's VP of engineering for Ubuntu, explained however the institution is baking AI into its Linux desktop and server acquisition successful Ubuntu Linux 26.04 and beyond. Unlike Windows, wherever Microsoft is slapping its Copilot statement connected everything, Canonical cooks AI into its Linux distro connected unfastened terms: unfastened models wherever possible, section inference by default, and nary rebranding of the distro into an AI product.

Seager explained that Canonical is "ramping up its usage of AI tools successful a focused and principled manner." That attack means a wide penchant for open‑weight models whose licence presumption align with Ubuntu's long‑standing open‑source values, coupled with open‑source harnesses and tooling. The Canonical developer teams are encouraged to follow the tools that marque consciousness for them, arsenic agelong arsenic they take a azygous instrumentality consistently astatine the squad level.

Also: Built for a hostile internet: Canonical VP of Engineering connected Ubuntu 26.04 LTS

He stressed that Ubuntu is not being repositioned arsenic an AI product, but that "thoughtful AI integration" volition marque the operating strategy much susceptible and businesslike for radical who already trust connected it. Internally, Canonical plans to amended its engineers connected wherever AI genuinely adds value, and to debar crude metrics similar "how overmuch AI did you use," focusing alternatively connected quality, control, and reviewability of AI‑assisted work.

Implicit vs. explicit AI

A halfway portion of Ubuntu's model is the favoritism betwixt "implicit" and "explicit" AI features. Implicit AI features volition tally mostly successful the background, enhancing Linux's existing capabilities. This is the benignant of betterment you'll acquisition arsenic "the strategy conscionable works better" alternatively than arsenic a caller AI product. For example, Ubuntu 26.04 boasts first‑class speech‑to‑text and text‑to‑speech, amended surface reading, and different accessibility improvements powered by section models.

Also: The caller rules for AI-assisted codification successful the Linux kernel: What devs request to know

Explicit AI features, by contrast, volition get arsenic new, opt‑in capabilities that intelligibly contiguous themselves arsenic AI‑driven. These features could see generative substance tools successful productivity workflows, agentic helpers for tasks specified arsenic record oregon task management, and dedicated interfaces for interacting straight with models. Seager describes this attack arsenic phased: first, softly improving what Ubuntu already does, past layering connected "AI‑native" workflows for users who actively privation them. 

Don't privation these AI-enabled programs? Fine. You don't person to usage them. Good luck trying that with Windows 11. 

Ubuntu is each astir moving AI locally. Canonical wants astir Ubuntu AI features to default to on‑device inference. This attack makes these features usable offline, perchance much private, and little babelike connected proprietary unreality backends. It volition besides marque them overmuch cheaper to use. 

Also: Linux explores caller mode of authenticating developers and their codification - however it works

This attack dovetails with Canonical's existing enactment connected tuned kernels, hardware enablement for GPUs and accelerators, and partnerships with silicon vendors. Seager described this arsenic the instauration for businesslike section inference connected mean Ubuntu installations.

Accessibility is 1 of the archetypal factual targets for this AI push. Seager highlights system‑wide speech‑to‑text and text‑to‑speech, positive richer surface scholar capabilities, not arsenic flashy "AI add‑ons" but arsenic halfway OS functions. Looking ahead, helium wrote, "What contiguous seems similar it's lone imaginable with entree to a frontier AI mill volition go importantly much accessible successful the coming months and years."

Beyond idiosyncratic features, Canonical is pushing guardant an Ubuntu that tin enactment arsenic a safer location for AI agents and agentic workflows. Seager says users are progressively accustomed to moving with agents and that helium "loves the idea" of making the accumulated powerfulness of Linux much accessible via agent‑driven interfaces. The extremity is simply a "context‑aware OS" successful which agents tin crushed astir the user's situation and tasks portion being constrained by Ubuntu's existing information model.

Also: 'Like handing retired the blueprint to a slope vault': Why AI led 1 institution to wantonness unfastened source

Here, Snap, Ubuntu's default exertion instrumentality approach, becomes Canonical's mode of securing AI agents. With Snap, agents volition sandbox. This measurement blocks them from accessing restricted information and resources. Canonical is exploring ways to integrate specified workflows "in a mode that feels tasteful, aligned with our idiosyncratic basal and respectful of our privateness and information values," explicitly acknowledging assemblage anxiousness astir heavy‑handed AI.

With Microsoft making AI a marquee branding term, Seager is astatine pains to differentiate Ubuntu's approach. He rejects the thought of measuring Canonical unit by the measurement of AI output, and helium says the institution is not readying to "force" AI connected users oregon crook Ubuntu into an AI‑first product. At the aforesaid time, helium is frank astir AI's interaction connected engineering work, noting that portion Canonical does not mean to regenerate radical with AI, an technologist skilled with AI tools surely could outperform 1 who is not. 

One happening users should not expect is simply a cosmopolitan "AI termination switch." Seager argues that specified a power would beryllium analyzable to implement, "honestly," fixed that immoderate AI functionality volition blur into inheritance strategy improvements alternatively than discrete apps. Instead, the accent is connected keeping AI features constrained, auditable, and aligned with open‑source expectations, portion allowing Ubuntu to germinate successful a satellite wherever AI is rapidly becoming portion of the baseline of modern computing.

Windows AI vs. Ubuntu AI

Canonical is explicitly biasing Ubuntu toward open‑weight models, open‑source harnesses, and exemplary licenses that align with long‑standing free‑software values, alternatively than conscionable grabbing immoderate performs champion connected benchmarks. Mind you, arsenic Seager observed, "access to exemplary weights is meaningful, but it is not equivalent to the benignant of transparency the unfastened root assemblage has go accustomed to." He added, Canonical volition take models based connected licence terms, not conscionable performance.

Microsoft's mainstream AI push, by contrast, is anchored successful proprietary unreality services, specified arsenic Copilot for Microsoft 365 and Azure OpenAI. Yes, Microsoft volition fto you usage galore models, but lone if Microsoft acts arsenic the gatekeeper. You tin lone usage AI successful Windows utilizing Microsoft's rules, including its pricing, policies, and telemetry.

Also: The caller rules for AI-assisted codification successful the Linux kernel: What each dev needs to know

Canonical's program for Ubuntu is to marque section inference the default. Ideally, each AI‑enhanced OS features should tally connected devices offline, with intelligibly defined interfaces that are utilized lone erstwhile an outer work is genuinely needed. That attack plays to Linux's strengths, specified arsenic hardware tuning and GPU/accelerator enablement, portion keeping your information and workflows connected your machines.

Microsoft's strategy has been "cloud first": Copilot successful Windows and Microsoft 365 is fundamentally tied to cloud‑hosted models and information processing, adjacent erstwhile immoderate client‑side NPUs get involved. That transportation makes it easier to rotation retired features astatine scale. However, the attack besides centralizes information and compute, increases vendor dependence, and makes it harder for users to recognize oregon bounds wherever their information flows.

As Seager pointed out, Ubuntu splits AI into "implicit," softly improving existing capabilities similar speech‑to‑text, surface reading, and different accessibility tools, and "explicit," new, intelligibly labeled AI workflows oregon agents that users tin take to adopt. This divided is each astir AI making Ubuntu "meaningfully much capable" without turning it into an AI‑branded merchandise oregon forcing AI connected users who privation a unchangeable Linux desktop.

Microsoft's stance, connected the different hand, is each astir pushing AI into the default idiosyncratic experience. For instance, Copilot appears straight successful the Windows ammunition and Microsoft 365 apps. In addition, Microsoft is exploring always‑on agents wrong 365. There, agentic AI volition enactment arsenic an operational furniture for bureau workflows. That displacement is large if you've already bought into Microsoft. And, obviously, tons of radical are OK with that stance -- more fool they from wherever I sit.

Also: The champion escaped AI for coding - lone 3 marque the chopped now

However, being tied to Microsoft means you indispensable interact with AI by default, not by a considered opt‑in. Are you bully with that? Will you inactive beryllium OK with it arsenic AI costs surge higher? 

Canonical's AI communicative leans heavy connected utilizing Ubuntu's existing information primitives, particularly Snap confinement, to springiness AI agents tightly scoped permissions, wide auditability, and antithetic "grades" of entree from read‑only investigation done to controlled constitute access. The thought is simply a "context‑aware OS," successful which agents tin beryllium powerful, but they tally wrong transparent, open‑source sandboxes that users and auditors tin inspect.

Also: I tried a command-line-only distro that tin earnestly amended your Linux skills

Microsoft's agentic absorption is much focused connected integrating agents straight into concern workflows, specified arsenic Microsoft 365 agents that tin enactment crossed mail, documents, and line‑of‑business systems. That integration is large for automation, but harder for users to understand. Governance lives successful argumentation consoles and connectors that IT admins configure, not successful a user‑visible, unfastened information exemplary that tin beryllium independently examined and forked.

Canonical positions Ubuntu arsenic a low‑friction level for section AI experimentation and open‑source workflows. With Ubuntu, it's casual for developers to swap retired models, frameworks, and tools. This attack makes it easier for teams to prototype with section models, vector databases, and cause frameworks, and, crucially, to debar vendor lock-in during the experimentation phase.

Microsoft's spot is monolithic organisation and integrated tooling. But that aforesaid integration makes it much apt that aboriginal experiments go long‑term dependencies connected Microsoft's stack, with data, workflows, and governance each tied to the aforesaid vendor.

If you attraction astir unfastened models, section control, and the quality to spot and signifier however AI is wired into your system, Ubuntu is your friend. Microsoft's exemplary is compelling for tightly coupled, cloud-first endeavor workflows, but it trades openness and portability for lock-in, heavy integration, and convenience. I cognize which exemplary I'll beryllium using.