About a year ago, a well-known financial institution posted a job ad on LinkedIn. The role: a generative AI expert with 15 years of experience. There’s one problem with that ad. The person doesn’t exist. Nobody on earth has 15 years of generative AI experience, and the fact that a major institution went looking for one says a lot about where most organizations sit right now: under pressure to do something with AI in the workplace, without a clear picture of what that something should be.

Dr. Hossein Rahnama has spent his career building that picture. He’s a professor at Toronto Metropolitan University, a visiting professor at the MIT Media Lab and the founder of AI company Flybits, with more than 50 published papers and 16 patents behind him. 

He joined Jordan Claes on Employment Hero’s Inside the Workroom podcast for a conversation that ranged from agentic AI to boardroom governance, and it was the LinkedIn ad above that he offered as the perfect symbol of how businesses get this wrong.

Here are the ten lessons we took away.

1. Start with one question: AI to do what?

Rahnama’s first move in any AI conversation is to slow it down. Under the umbrella of machine learning alone, he points out, there are 60 or 70 different ways to use AI. So before a single tool gets procured, leaders need to answer three things: what problem are we solving, what type of AI suits it and what does our own organization’s ecosystem allow us to do well?

That last part matters most. Companies have a DNA, in his view, and the ones that copy a competitor’s playbook instead of playing in their own league tend to fail. Figure out your genuine competitive advantage first, whether that’s your market, your people or a niche capability, then use AI to sharpen it.

2. Agentic AI has been around since 1994

The technology behind AI agents dates back more than 30 years. What’s new, Rahnama explains, is the ecosystem around it: enormous volumes of data, cheap computers and millions of people training and updating models on a daily, if not hourly, basis.

That distinction should change how leaders read the hype. The breakthrough you’re being sold on is really an infrastructure story, which means the winners will be the organizations with the best data foundations, and that’s something you can start building today.

3. Decide upfront: decision maker or decision supporter

Before deploying any agent, Rahnama says there’s a fork in the road. Do you want the machine to make decisions, or to support the humans who do? Neither is wrong. Alerting, monitoring and defence systems genuinely need AI as a decision-maker.

But for most workplace applications, he champions the human-in-the-loop model, where people stay in charge of process, auditability, accountability and outcomes while AI handles the heavy repetition. The point, in his words, is a workforce “being augmented with AI rather than being replaced by AI.” Choosing your lane deliberately, rather than by default, is where good governance begins.

4. Data-heavy, regulated industries will change first

Which jobs shift first? Follow the data. Any business built on repetitive historical data is prone to change, and Rahnama names law, medicine, finance and insurance as the front of the queue, since regulation is essentially a guardrail built from proven past experience, and that’s exactly what AI models digest best.

He calls today’s tools “screen AI” because they live behind our two-dimensional glass screens. Work that depends on hands, buildings and physical objects will still benefit from process efficiencies, but reliable AI in three dimensions remains a way off. If your business runs on documents and precedent, your timeline is shorter than you think.

5. Perspective-aware AI is already working

Ten years ago, Rahnama’s research into perspective-aware AI was theory. Today it’s running. The idea: turn trusted individuals’ expertise into models you can consult, so while reading a document you could ask what a colleague, a specialist or a mentor might think of it, without them being in the room.

His team has applied it in education, in healthcare and now in the future of work, including a digital twin of an office where a CEO can view the day through an engineer’s lens or an ESG lead’s lens. The payoff is context. Walk into a 15-minute meeting already understanding your colleague’s perspective, and you get the value of an hour without spending it.

6. Human skills are the durable advantage

At Flybits, Rahnama hires plenty of co-op students, and he’s watched the gap between a veteran engineer with 20 years of experience and a recent graduate shrink fast. Technical skills are becoming table stakes because the tools keep everyone within a standard.

What he screens for now is the ability to learn and unlearn, plus the human skills AI can’t automate: running a meeting well, being punctual, turning effort into results. He puts it plainly: AI skills will automate themselves, human skills won’t. For anyone rethinking their hiring criteria or development budget, that’s the signal.

7. Treat data as an asset class or accumulate risk

Rahnama’s biggest worry isn’t the AI itself. It’s the cyber and privacy risk companies quietly build up when they treat AI as an IT purchase, feed their data into someone else’s software and tick the box marked done.

Data, he argues, is an asset class. It has to be translated into information, contextualized, turned into wisdom and only then into action, and every step carries legal weight. Consumer data, corporate data, public data and government data all come with different privacy obligations. Companies that skip this thinking aren’t saving time. They’re storing up trouble.

8. AI transformation runs through HR

Here’s the answer that stopped our host mid-sentence. In the digital transformation era, the CTO or chief digital officer held the most important seat. Ask Rahnama who matters most in AI transformation, and his answer is the head of HR.

His reasoning: digital transformation was about connecting systems and APIs. AI transformation adds regulation, uncertainty, auditability and compliance, all at once, and the deciding factor is whether you have the right people with the right attitude to build in a new way. Technology you can buy. The talent and culture around it you have to grow, and that’s an HR mandate.

9. Swap explainable for auditable, and pilots for a living lab

Two terms Rahnama would retire. First, explainable AI: try explaining backpropagation to a lawyer. He prefers auditable AI, where any recommendation can be traced to the attributes, weights and data provenance behind it, ideally overseen by a dedicated board committee. If a customer calls to ask why your system sent them something, someone should be able to answer.

Second, the pilot. Most companies suffer from what he calls pilot exhaustion, running proof of concept after proof of concept that tests technology and nothing else. His alternative is a living lab: people, process, product and platform tested together as one machine before any big decision gets made.

10. Get comfortable not having the answers

Asked what one assumption he’d banish from a Canadian CEO’s mind, Rahnama took some blame as a professor first. Education has trained generations to be penalized for not having the answer, and that instinct is now holding leaders back.

His advice: “You should be very comfortable not having the answers, but be as curious as possible to ask good questions.” No consultancy, university or think tank knows everything about AI, and anyone claiming otherwise is selling a recipe for failure. Build a failsafe environment instead, where the vision never fails, but the team can afford micro failures within guardrails on the way there.

Where AI in the workplace goes from here

The thread running through the whole conversation is a shift in posture. The organizations that win with AI in the workplace won’t be the ones that bought the most software or hired the mythical expert with 15 years of generative AI experience. They’ll be the ones that knew their own DNA, treated their data with respect, put HR at the centre of the change and led with curiosity instead of certainty.

Ready to see how smarter employment software supports a growing Canadian team?