The Weekly Brief — Issue 06

Hello,

Quebec is starting to put infrastructure underneath a much bigger idea about AI in government.

On Friday, its digital ministry described the emergence of an “agentic state” — a future in which some routine government work could be handled by AI systems that can plan and take actions, not just answer questions.

That language is new. The infrastructure beneath it is already moving.

Quebec has authorized a $170.4-million execution phase for a shared digital-development platform that includes AI-assisted coding and future AI foundations in the province's government cloud.

Plus: AMD is buying a Toronto AI-chip startup, Shopify is opening more of commerce to AI agents, and a Kitchener-Waterloo company wants AI agents to do research themselves.

here is the week.

What happened

On August 7, Quebec's Ministry of Cybersecurity and Digital Technology updated its public-sector technology outlook with a striking phrase: the emergence of an “agentic state.”

The ministry says agentic AI systems can plan actions, interact with applications, analyze data and carry out defined operations. It lists request processing, document validation, file management, regulatory monitoring and operational support among the tasks that could eventually be automated.

That is still a statement about direction, not an announcement that Quebec has deployed autonomous AI across government.

But separate government documents show that some of the technical groundwork is already authorized.

In June, Quebec formally approved the execution phase of the Plateforme de développement moderne pour la transformation numérique, or PDM. The $170.4-million execution phase is intended to create shared development, hosting and support services for public bodies, and the decree requires the work to be completed no later than June 10, 2031.

The project's business case explicitly includes integrating AI tools into application development. Its development-tool service includes an AI-based coding assistant, and a later phase calls for deploying the platform's “AI foundations” in Quebec's government cloud beginning in December 2026.

The key distinction: Quebec has authorized a $170.4-million digital-platform project that includes AI tools and future AI foundations. It has not authorized a $170-million “AI project,” and it has not deployed an autonomous “agentic government.”

Why it matters

These are separate government documents, and Quebec does not say the PDM itself is an “agentic state” program.

The connection is more practical than that.

The PDM is designed to give public bodies common infrastructure, development tools and cloud environments. At the same time, Quebec is describing a future in which AI moves from experimentation into day-to-day operations and, eventually, performs defined tasks.

That matters because AI that can act creates a different set of public-sector questions than AI that only gives advice.

A chatbot might explain how to process a request. An agent could potentially interact with systems and help carry out parts of the process.

Who decides what it can access? What actions can it take without asking? What record does it leave? When must a person intervene? And how can someone challenge an error?

Quebec's own outlook says agentic capabilities should rely on human supervision, traceable decisions and rigorous protection of personal information.

The PDM also puts emphasis on portability, open technologies and the ability to move applications between government and public-cloud environments. The stated goal is to reduce technological lock-in and give the administration more control over its infrastructure and technology choices.

Your move

For Quebec residents, this does not mean AI agents are suddenly making government decisions about you.

The thing to watch is implementation.

As AI moves into real public-service workflows, look for clear answers to four questions:

What is the system allowed to do?
What information can it access?
When does a human have to review its work?
How can a person challenge or correct the result?

For governments elsewhere in Canada, Quebec is becoming a useful case study in a question many public organizations will face: what infrastructure and governance need to exist before AI agents are allowed to do real work?

AMD is buying Toronto AI-chip startup Taalas

AMD has agreed to acquire Toronto-based Taalas, a company founded in 2023 that builds specialized silicon for AI inference — the process of running trained AI models to produce results.

AMD says it plans to integrate Taalas technology into its accelerator roadmap and combine it with AMD Instinct GPUs. The company also says Taalas has a Canada-based team and that the acquisition reflects a commitment to retain and grow Canadian talent.

Why it matters: The deal is a strong validation of Canadian AI-chip engineering. It also raises a longer-term question for Canada: how much of the durable economic value created by homegrown AI companies remains anchored here as they scale? That is a strategic question, not evidence that Taalas is leaving Canada.

Shopify opens more of commerce to AI agents

Shopify reported 34 per cent year-over-year revenue growth in its second quarter, reaching US$3.58 billion.

Separately, the company has been opening more of its commerce infrastructure to AI developers. In June, Shopify made its agentic-commerce layer self-serve for developers. Its Universal Commerce Protocol and Catalog tools support experiences that can move from product discovery toward checkout.

Why it matters: One of Canada's largest technology companies is treating AI agents as part of the infrastructure of commerce, not just as an assistant inside a dashboard. The next question is whether merchants and shoppers actually get better outcomes — and how much control people retain as software takes on more of the shopping process.

A Canadian startup wants AI agents to do the research

Kitchener-Waterloo-based Transformer Lab has opened access to Primus, a research system built around multiple AI agents. BetaKit reports that Primus projects can use dozens or hundreds of agents running different AI models for different tasks.

The company says Primus can move from a research question through literature review, coding, experiments and a completed paper with limited human direction. It also says the system produced more than 30 papers during a 30-day run.

Those performance claims need qualification. Transformer Lab describes the papers as “Masters to PhD-level,” but BetaKit reports that the company did not explain how it assessed them at that level.

Why it matters: Primus pushes agentic AI beyond email, search and office tasks into the research process itself. If systems like this become reliable, they could change how research is conducted. But producing a paper is not the same thing as producing trustworthy science.

Most AI tools wait for you to ask a question and then produce an answer.

An AI agent can go further.

You give it a goal, and the system may break the job into steps, choose tools, retrieve information, interact with other software and take actions before returning to you.

Think of the difference this way:

A chatbot might tell you how to reschedule an appointment.

An agent might check the scheduling system, find available times and prepare the change for you — or, if it has permission, make the change itself.

That ability to act is why agentic systems can be useful. It is also why permissions, oversight and traceability matter much more.

The next time you use an AI feature that can take actions, check four things:

What can it access?
What can it change?
When does it have to ask me first?
Can I see what it did afterward?

The more authority an AI system has, the more important those answers become.

A calm, practical briefing on the AI developments that matter in Canada—delivered every Tuesday.

That's the week. See you next Tuesday.

— Padge T.

AI Brief Canada is an editorially independent publication published by Groundshift Advisory Inc. and written by Padge T.

Groundshift Advisory helps Canadian organizations adopt AI responsibly. Sponsored, affiliated, or commercially connected content will always be clearly disclosed.

Issue disclosure: Nothing in this issue is sponsored, affiliated, paid, or commercially connected.

AI use: AI tools supported research, source discovery, outlining and drafting for this issue. All material was reviewed, edited and fact-checked by Padge T. before publication. Read our AI Use Disclosure.

Not affiliated with the Government of Canada.

Corrections or feedback: We welcome corrections and reader feedback. Email [email protected] or review our Editorial Policy.