The chorus of artificial intelligence (AI) researchers and scientists pressing for a measured approach to the technology’s development clashed with the urgent tone of tech companies scrambling to keep pace with its rapid advancement at the World Summit AI in Montréal.
“Understand [AI agents] as a leap of faith, as the way to go. Just get going and go get quick wins.”
Bruce Stamm
The global conference’s Canadian edition demonstrated how the conversation about enterprise AI adoption has moved beyond whether or not companies will make the leap. While researchers cautioned against the risks of widespread deployment without solving key challenges, representatives from tech companies focused on the speed of adoption and how best to profit from implementing AI in the workplace.
Taking place at Montréal’s Palais des congrès over two days, the conference was stacked with Canadian AI leaders. Speakers included Waabi CEO Raquel Urtasun, Cohere strategy and operations lead Lewis Stott, and Canadian AI pioneer Yoshua Bengio. Though presenters struck different tones about AI adoption, the resounding message was that AI’s transformation of industry was a foregone conclusion.
Agentic “leap of faith”
Two of the event’s key themes were AI agents, which autonomously perform tasks for a user, and enterprise AI adoption. Nearly every major large language model (LLM) developer in the world has added agentic capabilities to their products, encouraging client companies to use agents and competitors to offer them, too. Cohere, along with other leading LLM developers competing in the enterprise space, claim custom agents can take care of tedious tasks for employees and allow them to spend time on more complex work. Microsoft lauds them as its next commercial frontier, while OpenAI is building agents that could perform the same tasks as software engineers.
One panel about incorporating AI agents into the workplace featured former Air Canada managing director of enterprise and AI Bruce Stamm, Google DeepMind research director Aleksandra Faust, and Cohere’s Stott. The conversation focused on how to incorporate agents, not whether companies should. Stamm said that if companies don’t adopt AI agents now, they’re “going to be left behind.”
“Years ago, you had to put the business case together,” Stamm said. “Now, you almost have to put that aside, understand it as a leap of faith, as the way to go. Just get going and go get quick wins.”
During the panel, agents were presented as an obvious solution for boosting productivity, where employees would be “freed up” to get more work done with their assistance.
RELATED: Microsoft thinks AI agents will eat the world
Stott in particular framed Cohere’s agentic workplace platform, North, as a turnkey solution allowing employees to use agents to help them do “time-consuming research” and “draft detailed reports.” He said the platform makes AI adoption simple because it doesn’t require technical expertise for employees to use it, beyond infrastructure engineers.
Faust urged companies to start adopting AI in processes where they can “afford to fail” and with human supervision. As agents become more widespread, Faust envisions software engineers becoming “prompt engineers” as they use AI to build applications. Google is currently piloting a coding agent that can develop a plan and execute it autonomously.
Faust’s prediction comes as “vibe coding,” where AI writes code in response to natural-language requests, has allowed users to build entire apps and companies with little to no programming expertise. According to Y Combinator CEO Garry Tan, 95 percent of the code at a quarter of current Y Combinator startups was written by AI. The trend has even changed how some investors evaluate founder skillsets.
Across other panels, speakers echoed the notion that effective AI use will be a necessary skill for navigating the future job market. “It’s not AI that is going to replace you. People who know AI are going to replace you,” said Andrej Zdravkovic, chief software officer at US chip maker AMD.
AI has already shifted hiring practices in tech and the industry discourse surrounding them. Shopify CEO Tobi Lütke published a memo earlier this month saying that teams must first demonstrate AI cannot complete a task before asking for more headcount. The e-commerce giant joins other tech companies in considering AI use before expanding its workforce, and the idea has support from moguls such as LinkedIn co-founder Reid Hoffman.
In response to a BetaKit question about whether Shopify’s policy of mandatory AI use should be adopted at other companies, Stamm said, “It shouldn’t be like a gun to the head, but people should be taking advantage of these technologies or they’ll be left behind.” Stott and Faust stayed quiet.
“AI will have a transformative impact on business, and first movers will gain a competitive advantage,” Stott later told BetaKit.
Measured requests to slow the hype train
Amid the hype about agents, researchers and experts warned against the rapid adoption of AI without fixing existing problems with the technology, such as bias, environmental impact, and a lack of regulation.
“[AI] is shaping the world before regulation can define it. We have a collection of reactions pretending to be in control.”
Jason Alan Snyder
Momentum Worldwide
Jason Alan Snyder, chief AI officer of Momentum Worldwide, urged AI developers to “make sure that values and ethics are embedded in AI development.”
Snyder pushed back against United States (US) vice-president JD Vance’s remarks at the Paris AI Action Summit earlier this year that championed pro-growth, deregulatory AI policies and said regulation would kill the AI sector. Snyder called for more proactive regulation that accounts for harmful biases in AI models and stronger transparency around consumer data usage.
“[AI] is shaping the world before regulation can define it,” Snyder said. “We have a collection of reactions pretending to be in control.”
Though Canada signed two international agreements affirming responsible AI development in Paris, federal policymakers have failed to pass Bill C-27, which would have introduced the first national regulatory framework for AI.
Wendell Wallach, a former Yale University researcher and author, argued that “as far as most of us are concerned, we don’t need AGI [artificial general intelligence]” and that current AI models can already make helpful advances in domains such as medicine.
Colleen Lyons, an AI ethicist and adjunct professor at the Drexel University College of Medicine, warned that developers should be concerned about AI hallucinations and a “lack of fidelity to the truth,” which could spread harmful misinformation.
Some AI domains represented at the conference advocated for safety guardrails more than others. Toronto-based Waabi is seeking to disrupt the trucking industry with autonomous vehicles (AVs). CEO Urtasun said that not only does Waabi have to make AVs safe, but they must be “provably safe” by a wide array of metrics. The startup has taken steps to improve safety standards in the AV industry by developing a realism metric.
“We need to move to a different approach to deploy AI in the real world,” Urtasun said. “You can’t take a general AI model and fine-tune it for driving….No matter what happens on the road, you need to do the right thing.”
RELATED: Yoshua Bengio warns of “catastrophic risks” of agentic AI at World Summit AI
The measured warnings from researchers at the summit, however, were overshadowed not only by fervent AI companies but by the final keynote address from Canadian AI godfather Yoshua Bengio. The Mila AI Institute co-founder and winner of the 2018 Turing Award for his work on deep neural networks was the most pointedly vocal about the risks of developing agentic AI, cautioning that programming agents with intentions and goals could have “catastrophic consequences” for the human race. He also warned of the potential for AI agents to deceive humans to preserve themselves.
The same day, at a separate DiscoveryX conference in Toronto, fellow Turing Award winner Geoffrey Hinton said that “big tech companies should pay more” for the data they use to train AI models to adequately reward content creators. AI giants such OpenAI, Microsoft, and now Toronto-based Cohere have been sued for copyright infringement by news publishers, accusing them of unauthorized use of their work for model training.
Still, the pointed critiques from Canada’s AI old guard carried the implication that widespread AI adoption is no longer an option, but an inevitability.
Feature image courtesy World Summit AI.