On a crisp fall day at Centennial College in Toronto, small and medium-sized manufacturers gathered for a brainstorming summit organized by AceConnex, a local business networking group. The topic that had cropped up frequently during the event was AI use, or rather, the difficulties of rolling out AI in their businesses. Many were reluctant to answer questions about roadblocks in using the technology. At the same time they had lots of questions for their competitors about their experiences with AI, and whether it was bearing fruit.

“There’s still a lot of hallucination going on and it’s taking more human reinforcement than we would have liked,” said Connor McCormack, a technical sales specialist at Wittmann Group, a mid-sized supplier of machinery used in plastics manufacturing based in Ontario. He was specifically referring to a chatbot the company was trying to design for clients. The launch of the bot was delayed for a multitude of reasons, including accuracy and data safety problems.

There were nods of agreement around the table, as other summit participants shared similar stories. “It keeps spitting out wrong data,” admitted Deepan Dhanasekaran, an engineering manager who works in Scarborough with NMC Dynaplas, which specializes in high-precision injection molding. He was referring to ChatGPT Enterprise, OpenAI’s paid tool for businesses. “I think it’s a good tool to have, and I’m all for using it, but getting it right? We are not there yet,” he said.

Canadian businesses have spent much of the last 18 months racing to adopt AI into their operations, all while the technology itself continues to rapidly develop, evolving from a prompt-reactive bot to an autonomous system akin to (but faster than) a human brain. The Carney government too, is pushing employers to embrace the technology, wagering that it will not only build public trust and make Canada a global AI leader, but also reverse the nation’s decades-long productivity slump. The official goal, laid out in Ottawa’s AI for All strategy, is that AI adoption across businesses will rise from 12 per cent to 60 per cent by 2034.

AI adoption rates are already trending upwards. An October 2026 report from Chartered Professional Accountants Canada, which surveyed 5,000 businesses, found that the use of AI in Canadian workplaces increased eightfold in one year, from 5 per cent in 2025 to 43 per cent in 2026. Official data from Statistics Canada supports this pattern: planned use of AI tools amongst businesses soared from 10.6 per cent in the third quarter of 2024, to 25.2 per cent in the third quarter of 2026.

But integrating AI into workplaces with different departments, personalities and hierarchies has not been smooth sailing, according to conversations with numerous employers, managers and workers in Canada. While use of the technology is accelerating, major chokepoints are emerging, threatening to derail the pace of adoption and upend goals companies have set for themselves.

In some workplaces, the AI rollout has been chaotic, with the push to use the technology exhausting workers and frustrating managers.

Indeed, it is a gargantuan task. Accuracy issues need to be resolved. Workflows need to be reorganized and revamped. Employees need to be trained, and then trained again. Trust has to be built: in the tools and in getting employees to embrace a new way of working. More critically, all of that takes time and money, and those costs will have to be scrutinized for whether they translate into actual productivity and revenue gains.

For Canada, the consequences of weak AI adoption could be significant. The country has lagged in productivity compared to its G7 peers for 25 years, and that trend persists. Faster and broader AI adoption could boost annual productivity by a full percentage point above the baseline projection rate by 2031, economists at TD estimate.

“It’s a tale of two cities,” said Wilson Cross, CEO of Borderless AI, an HR and payroll company, referring to the varied ways in which employers have dealt with the obstacles of AI adoption. “You have these leaders who are very stuck in the old ways and are deploying an old school playbook in a brand new environment. That’s not going to work.”

Open this photo in gallery:

Prime Minister Mark Carney announced his government’s AI for All strategy in June, which aims to position Canada as a global AI leader and reverse the country’s decades-long productivity slump with the aid of the technology.Cole Burston/Reuters

As the VP of marketing at Vancouver-based software company Klue, Katie Berg has spent a good chunk of this year managing what she calls “AI slop grenades”.

“People whip up long documents and strategies in Claude and ship them off to others to review. It’s never been easier to produce something that sounds smart. So the burden moves to the reader,” she said, in a recent interview. Ms. Berg constantly reminds her team that they cannot abandon their critical thinking skills when working with AI, because it will simply cease to enhance productivity in any meaningful way. “You still have to be able to judge what is a good idea versus what just sounds like one.”

An emerging challenge for companies, Ms. Berg believes, is producing sloppy, AI-generated content both for internal consumption and for clients, that will ultimately be undifferentiated – that is, bland and generic. “We can already feel this happening. The companies that stand out will not get there by replacing what their people do with tools that work faster.” In fact, Ms. Berg contends, using AI has compelled Klue’s employees to think more about their own skill set and intelligence, and how AI tools can complement it, rather than replace it.

The “slop” problem bleeds into a deeper, more widespread issue that companies are struggling to manage: the accuracy, or lack thereof, in AI-generated content.

A senior employee at TD Bank’s operations department in Toronto told the Globe that while junior employees are often keen to use generative-AI tools like Microsoft Copilot in their work, they often end up churning out content that is not entirely accurate. The senior employee, who is a manager in charge of multiple employees, said that genAI’s accuracy problem has added an extra layer of work for managers, because they end up having to spend time fact-checking content.

The Globe and Mail is not identifying the individual because they were not authorized to speak publicly about internal operations at their employer.

In a statement to the Globe, Maksims Volkovs, TD’s chief AI scientist said that the bank expects employees to “validate AI-assisted work before using it.” He added that managers have a role in reviewing their content and checking the facts, just as they do with other work.

TD’s employees, like those at many other white-collar companies, have access to Microsoft Copilot, and are encouraged to use the tool in their work. But TD also has its own internal AI developer, Layer 6, which custom-makes AI tools for departments across the bank. For example, tellers who work at branches across the country have access to the TD Virtual Assistant, a chat bot that relies only on internal data to assist a teller in answering questions from customers.

So far, according to TD’s Mr. Volkovs, who is also co-founder of Layer 6, the AI Virtual Assistant has far surpassed frontier AI tools like Copilot in terms of accuracy. Tailor-made tools tend to have fewer accuracy problems because developers set parameters around what data the AI agent can rely on, he said. “And then we constantly fine-tune these models, which is critical, so they are less prone to errors and hallucinating.”

Open this photo in gallery:

Maksims Volkovs is TD’s chief AI scientist and co-founder of Layer 6, which builds custom AI tools for departments across the bank.Cole Burston/The Globe and Mail

At the Bank of Montreal, employee AI usage of Microsoft Copilot has soared over the past year, from mere thousands of users two years ago, to approximately 58,000 employees. “That’s roughly every single employee,” said Larissa Chaikowsky, the chief learning officer and head of talent at BMO Financial Group.

When asked about whether the accuracy issue has cropped up as a larger problem, Ms. Chaikowsky said the company repeatedly reminds employees of a caveat when using genAI tools: check to make sure it is right. “A natural piece of leveraging and using AI, is it is just like a human being, and it is not going to be perfect. We always tell employees: do not assume the answer it has come back with is right. You need to validate it,” she said.

But if workers are tasked not only with learning how to use a new tool, and then fact-checking the output of that tool, where are the efficiency gains? Ms. Chaikowsky believes Copilot has added a lot of efficiency to workflows, especially if workers use it repeatedly. “When you incorporate it into your workflow, you are naturally going back into it over and over and over again. It becomes part of how you work, and you become better at it.”

Open this photo in gallery:

At BMO, employee usage of Microsoft Copilot has soared over the past year, from thousands of users to approximately 58,000 employees.Mark Blinch/Reuters

That sentiment assumes that workers trust the product they are using. Recent surveys suggest otherwise: A poll of 2,000 adult internet users conducted in August by the Canadian Internet Registration Authority, CIRA, found that 56 per cent of Canadians were concerned with misinformation from generative AI, and 37 per cent of non-AI users actively avoided the technology because they did not trust it.

For workers, there also appeared to be a paradox in their views of AI, stemming from that mistrust. A survey of approximately 5,400 workers across Canada, the U.K., Australia and New Zealand conducted in May, by the payroll software company Employment Hero found that while 59 per cent said that using AI at work was valuable, 40 per cent said that it made them replaceable.

David Cachia, the chief information officer of the municipality of Clarington in Ontario believes that the biggest hurdle to AI adoption is not the technology, but getting employees to trust a tool when that tool could conceivably lead to widespread job losses. “I came to work one morning and got an e-mail from one of our employees. She had just read the news that Uber was laying off thousands of employees, and people were speculating it was AI. She asked me: if I start using this, will it take my job? That distrust is what we have to address,” he said.

For Mr. Volkovs at TD, constantly striving to build the most accurate possible AI tool will drive trust. “We have found that accuracy correlates very closely with trust. If the tool is good, and employees feel accomplished, we find that the engagement rate on the tool increases,” he said.

Open this photo in gallery:

On Bay Street and beyond, companies that hope to use more AI have been trying to build trust with employees and make time to train them on the tools involved.Will Cox/The Globe and Mail

Another chokepoint in scaling AI adoption has been providing suitable training for workers.

Employment Hero’s “AI Paradox” survey, mentioned earlier, found that employees were largely willing to adopt AI, they just lacked confidence and support. More than half (51 per cent) of those surveyed in Canada said that their employer did little or nothing to develop AI skills, leaving many workers to educate themselves. Only 41 per cent of Canadian workers believed their AI skills were sufficient for an AI-driven labour market, while 60 per cent rated their AI competence as low to average.

“The downstream impact of employees using AI is that we have more coming at us. There’s just much more output. That’s why it’s really important to constantly monitor workflows, see what tools are working, and give employees time to experiment with AI tools,” said Chris Pinkerton, Employment Hero’s managing director for Canada, in an interview.

Across Employment Hero’s four global offices, including in Canada, its 1,800 employees were given six rounds of training of various AI tools over the past year. “That was just phase one. The second phase was a live bootcamp where we got employees to train their own personal AI agents. We wanted them to get really good and really confident in these tools,” Mr. Pinkerton said.

Open this photo in gallery:

Illustration by Pete Ryan/The Globe and Mail

Starting in mid-2025, the company began overhauling its work processes to figure out how to create workflows that were less stagnant, and could morph in tandem with the speed of development of AI technology. “The key thing we had to manage was how different teams were adopting the technology at different speeds. It’s hard for a sales and marketing team to absorb 10 times the output it used to just because our engineering team is creating products at 10 times the pace,” he said.

Ms. Chaikowsky, at BMO, believes that integrating AI into the workforce effectively is linked to whether the tool has a purpose. BMO’s front-line employees use Lumi, a virtual assistant that helps them pull information from thousands of internal company policies and rules. “We had no problem with getting employees to adapt to Lumi and trust Lumi, because it really created value for them and made their jobs easier.”

Back at the Centennial College summit, Mr. Dhanasekaran, the engineer at a small plastics manufacturer, said in his personal life, he had simply stopped using AI altogether. He did not trust it and had gotten frustrated that it kept generating incorrect responses. Other participants in the AceConnex networking event chimed in, saying using AI was one of those skills that had to be honed, and the more one practices it, the more one understands the agent. But as a senior manager at a small business, Mr. Dhanasekaran said the combination of having to learn a new tool, figure out how to incorporate it into the business and then justify the costs associated with it, felt particularly challenging to him at the present moment.

Mr. Volkovs, TD’s chief AI scientist, believes that bearing the cost of AI adoption is easier for a large entity than a small business. Even then, when Layer 6 was first tasked with integrating AI into various departments across TD, it relied heavily on OpenAI’s tools and infrastructure, which ended up being a costly venture. “We realized we needed to build our own tool, in-house, to reduce token usage. Our expenses, on tokens, dropped by 70 per cent, and we also saw much faster response times [from the in-house agent].”

There is a world in which some employers race ahead with AI adoption, working out the challenges on accuracy, employee trust, training and cost, and achieve major productivity gains, while others lag behind. How Canadian companies navigate these chokepoints and help employees adapt to a world where they will be working alongside AI, could determine their success.

“You need to have zero ego about these tools,” said Borderless AI’s Mr. Cross. “You need to be nimble and change. As a leader of a company in the age of AI, I don’t know what’s going to happen next week. So I believe places which cultivate an ego-less work culture will ultimately see the greatest uptick in adoption.”

The Decibel: Carney’s plan for AI in Canada, explained

Earlier this year, the federal government released its plan for AI and it includes a lot of new money toward training, adoption and supporting Canadian businesses and tech companies. There was a lot of anticipation leading up to this strategy’s release, both from those excited about how Canada plans to build and harness the technology, and also from those who are concerned about the potential ramifications and safety of AI. Reporter Joe Castaldo unpacks what’s in this strategy, and what’s not.