Every Canadian business owner has felt it by now. The pressure to do something with AI. Anything. Buy the software, run the pilot, tick the box before a competitor does.

That instinct, according to one of the country’s leading AI researchers, is exactly the problem. The businesses accumulating the most risk right now aren’t the ones moving slowly. They’re the ones treating AI as a procurement exercise.

Dr. Hossein Rahnama has spent two decades in this territory. A professor at Toronto Metropolitan University, visiting professor at the MIT Media Lab and founder of AI company Flybits, he holds 16 patents, has published more than 50 papers and heads the {sAIpien} program at MIT. In an interview with Employment Hero’s Inside the Workroom, he doesn’t name the technology as his chief concern. “The biggest worry that I have is literacy. AI literacy on what agentic AI is and is not, and what is the best way that we can use it in an organization.”

The industries built on old data will move first

Start with a fact that surprises most people: agentic AI was invented in 1994. The algorithms aren’t new. What’s new is everything around them: the data, the computing power, the millions of people refining these models by the hour.

So the question for employers isn’t whether the technology works. It’s what job to give it. Decision maker or decision supporter? Rahnama says neither answer is wrong, but the intent matters enormously. “We want the person, the individual, the human to be involved and being augmented with AI rather than being replaced by the AI.”

And where does that augmentation land first? Not where the tech headlines suggest. “Any business that is very reliant on data, especially repetitive historical data, is prone to change.” Law, because legislation is a frozen set of procedures. Medicine, insurance, finance, because regulation encodes decades of precedent into interpretable data. “I’m not saying necessarily negative disruption, but they are prone to change.”

An AI that tells you what a colleague would think

The more surprising thread in Rahnama’s research is what he calls perspective-aware AI. It’s a pointed departure from the chatbots most of us use, which he describes as conversational interfaces sitting on “billions and billions of parameters” that in many cases offer “the regurgitation of the past” with little visibility into where any of it came from.

His team is building something different: models of individual human expertise. Imagine reading a difficult document and being able to ask how a trusted colleague would see it, without that colleague being anywhere near the room. “You’re not telling me what to do. You’re not telling me if this is good or bad. You’re just sharing a perspective.”

They’ve tested it with professors examining a subject through different academic lenses, and with medical specialists reviewing the same patient file. Now the focus is the workplace, where a digital twin of an office lets a leader borrow an engineer’s view of the day, or check carbon metrics through the eyes of the person who owns them.

The practical payoff shows up somewhere every business owner will recognize: the calendar. “If I have an upcoming 15-minute meeting with you to get an update, I have a lot more perspective and context that I can make the most of that 15-minute meeting rather than turning that meeting into a one-hour meeting with the same level of performance and productivity coming out of that.” None of this replaces human conversation. It just means arriving prepared.

The Head of HR just became the most important person in the building

All of it, though, rests on data. And this is where Rahnama, with the caveat that he’s sharing a personal opinion, gets blunt about what companies are fumbling. “One thing is, companies do not treat their data as an asset class. Data is an asset, right?”

The familiar pattern goes like this. A CEO hears about AI at a conference. The enthusiasm trickles down to the CTO’s office, some software gets procured, data gets fed in, and leadership puts a check mark next to AI. What’s quietly piling up underneath, he warns, is cyber and privacy risk, because consumer data, corporate data and public data all carry different legal obligations that a software licence doesn’t resolve.

Which is why his answer to the question of who matters most in an AI transformation lands the way it does. “If you ask me who’s the most important person in AI transformation, it’s the head of HR. Because you need to make sure you have the right people with the right attitude to be able to build in a new way, and the old model doesn’t work.”

The people problem is real. He recalls a well-known financial institution posting an ad for a generative AI expert with 15 years of experience. Fifteen years. In a field this young, that person does not exist. “We are now entering an era that you need hundreds and hundreds of these talents to be available. Where are these folks? They are in K twelve.”

What to actually hire for

If the talent is still in school, what should employers look for right now? Rahnama’s own hiring at Flybits offers a preview, and it cuts against instinct. “The gap between a veteran software engineer with twenty years of experience and a recent graduate from an undergraduate computer science program is shrinking very quickly.”

Technical skill is becoming table stakes. The differentiator is elasticity. “What matters to me now is your ability to learn and unlearn.” The executive who built a career selling one way may need to discard that playbook entirely.

Then there’s the part no model can touch. Running a meeting well. Being punctual. Getting to a result. “It’s human skills, it’s not AI skills. AI will automate itself. It cannot automate human skills.”

Curiosity beats certainty

So where does a business actually begin? Not by copying anyone. “Companies have DNAs and this DNA, after the first few months of inception, will always stay with the company.” Find the genuine advantage, whether it’s the market, the people or a niche capability, and augment that. And the push has to come from the top, because bottom-up AI initiatives stall without leadership buy-in.

As for the endless pilots so many companies are running? “It will never work. Do you know why? Because the old notion of a POC and pilot in most cases is just about technology.”

His closing advice is the piece worth pinning to the wall, and he offers it with a professor’s self-awareness about an education system that punished wrong answers for generations. “You should be very comfortable not having the answers. But be as curious as possible to ask good questions.” No consultant, university or think tank knows everything about AI, and any that claims to is selling a recipe for failure.