Has Canada already lost the AI race?
It's not a race where there will be a single winner.
OK. Amazon, Google, etc. — what are they?
There are five of them sharing a cake that is 80% of the American stock market value.
Yes, but there are so many other companies. We could go to the top 1000 and we'd be happy to have three times as many in Canada.
So Dominique, but at some point the only way to lose is to stop running. It's not a sprint, it's a marathon, this thing. And the question is: are we going to get somewhere interesting? The only way is to keep going.
The Solidarity Fund FTQ is proud to present Les Dérangeants because we care about the Quebec of today and tomorrow. We are here to support local leaders who are committed, visionary and passionate. Together, we invest to have a positive impact on our society.
The podcast of the new generation of entrepreneurs in Quebec. Les Dérangeants.
Les Dérangeants. Hello everyone. Welcome to this new episode of Les Dérangeants. Étienne Crevier hosting, welcoming you from the En Ligne 2026 conference at the Palais des congrès here in Montreal, to talk about the hot topic that is artificial intelligence and whether this is the end of humanity. We're going to discover, actually, Les Dérangeants. What is it? You've known us for 9 seasons already. It's the podcast of the new generation of entrepreneurs, proudly presented by the Solidarity Fund FTQ, Kaa Capital, Blue HF, and several others we'll be announcing. Some great changes for season 9. Actually, to accompany me on this topic, as I said, is Canada — which is technically the inventor of artificial intelligence with our researchers, the Yoshua Bengios of this world among others — did we get bypassed in the end, and is it too late? So Dominique Gagnon, hello.
Listen, you who are a techno-pessimist, a bit of a pessimist — I'm an obsessive user. That means I use it a lot, but I'm mostly critical. You know, it's obviously — I'm starting to be known at Les Dérangeants as the person who criticizes and who shakes things up a bit. I mostly ask questions about things like the announcement — I think we announced this week a $300 million investment in artificial intelligence. I know that for the average person, we go 'wow, $300 million, that's a lot of money,' but it's nothing when you compare it, for example, to what's invested in the American market. And the question we were asking a bit today at this summit — it's still crazy to say there are 10,000 people right now at the Palais des congrès for En Ligne, all people in artificial intelligence. It's buzzing with talent, knowledge, startups. But a question I was asking is: are we now in the last peloton in the race to win the AI race? And we're going to receive, I think, the best person to talk about that, who is Joelle Pineau from Cohere — but for me, she's also the former Chief AI Officer of Facebook, of Meta, and today a luminary who was named among the 100 most influential people in artificial intelligence. So I'm quite, I don't know about you, honored to be able to receive a person like that on the podcast.
It's crazy, because we're never masters in our own house. We have plenty of Quebecers around the world — the CFO of Google is a Quebecer, the Chief AI Officer of Meta is a Quebecer — but we have talents, brains here that we don't know, that we don't showcase enough. And indeed, it was complicated — or rather not complicated, but we were honored to receive Joelle, that she accepted to take time during En Ligne, during her conference, to spend 45 minutes with us. So we'll be able to dig into this question, because you were saying you're an abusive user of artificial intelligence. Me, I still like to make the joke that ChatGPT is actually an army of Indians answering my questions on the other side and it doesn't really exist. What's your use of AI?
Ah well, for me it's a use where I'd say I have about twenty agents working for me in real time. Which means that while we're recording the podcast, tasks are getting done. If I have presentations to make, they might be reading the media for me and summarizing it. So I've really automated a lot of things. I use them daily to record meetings, to automate tasks, to read and sort my emails. So for me, it's a way to reclaim time. One thing entrepreneurs don't have enough of is time. We're always spending — per month, per year — I spend today maybe around $1,500 to $2,000 a month.
Wouldn't you rather have a human instead for that price? You could technically have an assistant. But there, you're talking about an assistant, which means that today with AI, I'd probably need to hire 10 or 12 people with different skills to do that work. I think you're bringing up the big problem of Canada very often — everyone talks about cost. Everyone tells me 'but how much are you spending?' Dominique, it's terrible, $1,500 a month. Nobody asks the question of what's my return on investment? How am I able to save time? Me, I think today I recover between 15 to 20 hours a week by having used artificial intelligence.
20 hours, Étienne. Let's say times $100 an hour. We're talking roughly $2,000 at that point — saving 20 hours a week times 4 weeks. We're talking about $10,000 that I can recover from that side — $8,000. Me, I'm not very good at math, definitely dyslexic, but we're talking about an enormous amount of money, actually. Paying $1,500 to recover $8,000 — who would say no to that? I think you have to invest, and it's one of the problems. It'll be part of the discussion I'll have with Joelle: to what extent Canada has the means for its ambitions, and to what extent we're investing enough to really be able to keep being in this AI race and not be only researchers, because that's a lot of what we see in Canada — it left us with a bit of a bitter taste, definitely.
But not just that, because you know this episode was filmed during En Ligne, so we're in September, we're in the middle of an election campaign, it airs later, but right now for me it's the big criticism I have for Quebec — that for example the Lion Electrics of this world, we let them die because it wouldn't have taken $300 million for Lion, it would have taken $3 billion to have a chance. And we didn't have access to that capital. There weren't people with an appetite for risk. Do we have it for AI in Canada, according to you?
Well you said it yourself — we're in the middle of an election campaign, we're talking very little about artificial intelligence in this election campaign, which is quite surprising. There are Quebecers who brought it up once.
Yeah, it's us and I think the Liberal Party. I know you're more Parti Québécois and me Liberal Party, doesn't matter which. I think everyone talks about it a bit but not enough given the important change it will bring to society. We also don't talk much about education — that will definitely be part of this important topic of education. But artificial intelligence today, for me, is at the center of a bit of everything, and you used the right example with Lion Electric, but we've also invested enormously in foreign companies. I think there's a change that will happen soon, but I agree, we'll have to have the means for our ambitions or otherwise think differently, maybe more in partnership. It'll be an interesting topic to talk about with Joelle too — the fact that they merged with a German company. So today, we really see that Canada wants to get closer and closer to Europe. I think there's something interesting in that partnership because we really often share values with Europeans.
So there, you're spending $1,500 to $2,000 a month on artificial intelligence. If you, Dominique Gagnon, had access tomorrow morning to a Canadian sovereign model but it would cost you $2,500 instead of $2,000, would you do it?
Unfortunately, I don't think so. Again, I think at a certain point you also have to understand that there's a price for everything, but it has to make sense. So if I pay $1,000 more per month as an individual — we're really talking about an individual choice of Dominique's — it's money out of my pocket, and the only reason is that it's a Canadian model. Me, I want Canada to be better and more efficient because we also have enormously positive points for artificial intelligence. Think about energy. There aren't many countries today that have a more strategic position than Canada on energy. So the reality is, would I be willing to pay a few hundred dollars more to encourage a Canadian company? Yes. What I want is Canadian companies that are also performant, that have at least the same price-quality ratio as our American neighbors. Can we rival China on price? That will definitely be a challenge. There are also risks to using Chinese models from a security standpoint, and from an ethics standpoint, I think. And for sure, if you offer me to use, for example, a Canadian AI tool and I pay double just because it's Canadian, well I don't really see the advantage.
That will be, I think, the big blocker, because we talk about digital sovereignty. Sovereignty isn't — as we said — it's not political, Quebec separating from Canada. AI sovereignty is having access to models that are independent of other countries' policies, of American policies. But the question — you can think of Fable, you know Fable, where Donald Trump decided one fine morning to say if you're not American, you don't have access anymore. There were businesses already running on that model, that had started, being told it's over, we're unplugging you. Imagine today if a government could tell us you no longer have access to the internet — the economy would collapse at such a fast speed. So for me, sovereignty is having options.
I would say it's optionality. It's the aspect of saying: I'm not stuck with Microsoft, I'm not stuck with Google, with Gemini, I'm not stuck with Anthropic — I can use a Canadian model, I can switch. And there's really the aspect of having a certain freedom of choice, freedom of use, and that nobody can decide for us whether we have the right to do something or to use a model in our productivity.
Well, let's dig into the question with Joelle Pineau, none other than the Chief AI Officer of Cohere, here in Montreal during the En Ligne conference, to answer this question: Has Canada already lost the AI race? We'll answer that right after the break. Joelle, hello, welcome to Les Dérangeants.
Hello, it's a pleasure to be here.
Listen, you're probably the best-placed person in the country to answer the episode's question we want to ask today, because you lived the AI race from the inside. Actually, you were born in Ottawa, you did your engineering at Waterloo, a PhD in robotics at Carnegie Mellon no less, computer science professor at McGill and member of Mila. You led FAIR, the AI lab of Meta Facebook. You were Chief AI Officer from 2017 to 2025, and since 2025 you're the Chief AI Officer of Cohere. You're a member of the Royal Society of Canada and you were named by Time Magazine among the 100 most influential AI people in 2026. Nothing less. What letters of noblesse are you missing?
We're not here to hunt for letters of noblesse. Listen, this is a fascinating era. And really the possibility of participating in the development, the blossoming of this technology, is truly a privilege but also a big challenge.
Well listen, I'd be curious because you've lived others with Meta — you were among the American giants. So the conference we're at today, according to you, what is it? The equivalent of the Quebec Games final, or are we at the FIFA World Cup championships?
We're a bit in between. There's definitely a big international presence this year. Germany, which is the country of honor, really — there's a concern to find a way to build alliances that allow us to keep developing our technology. We know the competition is fierce in this domain, and it moves very, very fast. So we have to work together. But I met American, German and other researchers I didn't expect to see here, and then suddenly they show up on a panel. So it's clear the event is on the map internationally, and from one year to the next it's growing. Really impressive to see the energy.
Ah me, Dominique, I don't know about you, but last year when we came to Les Dérangeants, it was much smaller. This year, there's been a step up — we're talking Quantum Computing, I saw intelligent drones, I saw the supercomputers from Waterloo. There's something this year.
Yeah. And enormous amounts of people. It's so difficult just to walk, sincerely, without getting stopped 287 times.
Yeah, it was quite intense on that side.
So the question we're bringing up a bit — and you know, I think the moment is also quite pivotal. We can't, I think, do a podcast without talking about the recent media statements from a former OpenAI person who keeps saying AI will destroy the planet, we're all going to die, and we should stop this race. I'd like to have your opinion before getting to Canada — is it really in the race? Because I'm part of the people who ask the question: are we trying to join a race in which the winner, or at minimum the winners, are already decided in advance? We talk everywhere all the time about the Chinese and the Americans, very little about Canadians except for research, you know. Obviously, we have Yoshua Bengio who puts us on the map enormously. You, I think, Cohere puts us on the map — we often have the impression we're very good at doing research, and when it's time to make money, well, we're a bit less present. So to start, what's your position on the recent statements saying we should slow down AI development, that we should take a pause, and on these real dangers? The listeners — my parents actually asked me the question, saying 'Dominique, we know you're very pro, but are you sure it's a good idea?'
It's not just yours and mine asking me the same question.
A more complete answer with you, no less.
Yes. [laughs]
And listen, it's certain that the development is very, very fast. And without giving a whole history lesson, it's fun to look back at the end of 2022, 2023 when ChatGPT came out. It was like there was a collective reset — suddenly everyone could go there. It was super simple, you could go chat with it. Everyone had an idea of what the technology was capable of and was able to have that experience at a very personal level. Since that moment — we're talking 3 and a half, almost 4 years — a gap has been created between those who stayed at that point and those who follow the progression of capabilities, the agents, multi-agent orchestration, the coding agents, all that. So a gap is created at the capability level. There are people on the acceleration curve who see this and then suddenly project it 10 years ahead. But where we are right now in being able to predict that with fidelity, with rigor — completely impossible. It's astrology in a certain sense. It's science fiction. Science fiction is super fun. We create catastrophe scenarios and with that it makes us think. It's important to think, but we shouldn't take it as really being science. So what science tells us is that capabilities are accelerating, that we need to develop alongside those capabilities ways — guardrails — to keep things safer. So being able to do both at the same time is hyper important. And in Montreal, we have the luxury of being able to do both. We have Cohere developing the technology, we have our universities, we have IVADO and others informing the safety side.
And actually for the average person, the guardrails — I'd be curious to hear. Does a kill switch exist? Can we aspire to see that, or what would these barriers look like that would allow us to say yes, AI is in a manageable sandbox?
And we talk about guardrails, but we're talking about a president south of the border who doesn't seem to have many guardrails. On the other side, he's super loose, and on the other side there's China, which isn't necessarily known for being first in terms of human rights and respect for rights. So I come back a bit to your sub-question, Étienne, saying in Canada sometimes my impression is we want to do things right. We want to do things as safely as possible, but if you have two cowboys on both sides saying 'well, I'm going to shoot everywhere,' are we still going to be in the race?
But it's interesting to see that. And the term 'kill switch' was used recently but in two different senses. On one hand, there are people who say 'we need to be able to really shut down that system for security reasons.' And there are others who say 'we could have our access to that system removed.' We saw the American government block access to Fable's models overnight. Suddenly, people who depended on the technology for their business model, their productivity, no longer had access to it. So this issue of control and access to the technology, but also developing it and deploying it safely, is essential. We need to have options. So when we talk about sovereignty at Cohere, we talk about it a lot. People find the concept a bit confusing. What is sovereignty? We're not talking about selection, we're talking about having control over that technology, controlling it. So on the security side, it's possible to do. We talk about sandboxes — sandbox, I'm not sure it's the best translation. Anyway, we'll launch that to the linguists at Radio-Canada, you're allowed to say sandbox. But listen, we're able to do it. And we do a lot of deployments that are really what we call on-prem. So it's on the system, it's a closed system, completely cut off from the outside world. And the AI agents actually go fetch all the local information of the company. That's hyper performant, hyper secure by design. So when we're in regulated sectors — banks, health — we can build systems like that. Obviously, there's a cost, you need your own computers, you can't run on the cloud. You can't see the same thing on your computer, your phone, and send it by email to your mother-in-law. So there are trade-offs to make between security and interoperability of systems, but these are choices that get made, and with this concept of sovereignty which is a notion of choice, these are choices that companies and individuals can make. It's a bit the pitch of Cohere in a certain sense, that by adopting Cohere, a company can have a Canadian model and the data isn't used to train an American giant like Claude, like OpenAI or others.
It's not used to train, period.
Yes. Yes. Because we deploy on the user's system, we don't see it, we don't even see your data.
That's necessary for some of our clients. It's something we respect. We built products that meet that need.
And I ask the question — you made the choice at the very beginning to go to the Americans. So you were at Meta. I don't remember the book anymore, but there's a book you read that makes you hate Meta, that's a bit about the inside of Meta and how — no, I know it wasn't you, but it's a book that talked about how much Meta, among others, made people addicted. There are a few, you know, who bring that up. You decided to develop AI there for a while, today you've come back — is it because you felt there weren't the guardrails, that at the American level we'd reached a bit of a madness and a somewhat more capitalistic race? Maybe also what we're doing there — because I understand well the concept you're explain, but the reality sounds more complicated, it sounds slower, it sounds longer and it sounds more expensive. And I always come back — maybe it's my impression as an entrepreneur. I have the impression — could an SME tomorrow morning install Cohere that way, or are they still obliged to use American AIs?
You had a lot of questions there. What was your kill switch with — but let me go back first.
Because I decided to go there in 2017, and at that time I'd been a professor at McGill for several years. For years I'd watched each of my students, one after another, head south of the border. Master's, PhD, all without exception. And then suddenly there was a spark in Montreal. It was like the American companies had realized we had this talent pipeline. And for years I received calls, and I always told them: if you're ready to open something in Montreal, call me, but otherwise it's not worth it. And Meta called me because they wanted to open something in Montreal. So the vision was to create a place to keep talent in the local ecosystem. And at some point, we have to make partnerships. At that time, it was great to be able to keep all that talent in our ecosystem — talent that we employ today at Cohere and other startups. So that was really the intention. And at Meta, it was the only company that not only was going to open a lab — there were others — but that was ready to do all its research in open source. So all the work we did over the 8 years I was there, almost all of it — we can talk about it — was in open source. Meaning any small startup could take it, could use the Llama models. So today, you're asking me the question — can we get something? You have Llama in open source, but the reason you have Chinese models in open source is because we opened the way to that. That was a super interesting conversation internally. What does it take to convince a leader like Mark Zuckerberg to give away his models for free when all the other competitors keep them closed? OpenAI, Google and others kept their models closed. So that was a lot of my mission during those years. There came a moment where I felt I had done what I had to do there after 8 years. It was time to do other things. It was time to take everything I'd learned there and apply it and try to grow a company.
If we had the solid foundations to dare to build in Canada.
The Cohere pitch to convince you to make the jump.
Uh, they didn't have to work very hard. [laughs] The pitch is really the company's mission. Listen, it's an incredible privilege at this point in my career to be able to take a few months and tell myself: what do I want to put my energy into right now? And Cohere, the only company right now that has the capacity to train its models outside the United States and China. That's it.
Meta was doing it at first but they abandoned it.
Yeah, I don't want to speak for their strategy, you'd have to ask them.
I'd be very interested to hear the answer.
But it's clear that, let's say, they have other parts of their strategy that are more obvious. Whereas at Cohere, it's the reason for being of the company and it's at the heart of the technological development. There are...
No others at the moment. If you had a magic wand, so that today Canada has all the tools to compete on the world stage — if we were able to convince governments, to convince industry to support Cohere — what would you change with your magic wand?
There are a couple of things. First, we're in an election campaign — that's a good moment. Listen, it's the investments. Venture capital continues to be very difficult here. Compared to elsewhere, it's the attitude toward risk-taking. The bulk of the capital is Canadian and foreign capital, if I'm not mistaken.
It's a good mix. It's a good Canadian mix. We'll have the Germans participating and the Americans. I couldn't tell you the proportions, but there's still Canadian ownership.
I won't get into the shareholding. Our headquarters is in Toronto, I'm a co-founder and all that. So it's absolutely a Canadian company by law, by culture, and by what we develop.
But yes, access to financing to train these models — the cost of the computers, all the infrastructure — is incredible. And even if you manage to get the financing, listen, we receive quotes and have to give an answer on enormous expenses within 48 hours because the market is so volatile. So it's the capital, it's the GPUs. After that, it's the talent — keeping researchers invested in that mission when they're being made fabulous offers by the American giants.
How much can someone earn at an American giant? A fabulous offer — are we talking 1 million a year, 500,000? I'm curious.
The numbers are there. The numbers are there. Yeah, yeah, yeah, yeah, yeah. It's football-player salaries.
OK, OK. When we talk about football players, it's surely not the Canadian league. [laughs]
It depends where they're located. But so all of that is difficult. And then once we've built those models, once we've built the products, we then have to get clients to take a chance on a small company competing against a Microsoft with years of experience deploying industry solutions, with all their sales apparatus. Building all that on the commercial side is an enormous challenge.
On top of that, we add the regulation aspect, which is absolutely necessary. We need regulations around AI — we know that — but they have to be done well. And doing good regulation when the technology changes so fast, you know, if we looked at our regulatory ideas from 3 years ago versus the technology now, we'd be completely off base. So being able to think it through well is a big challenge.
