I recently spoke to a room full of engineering students at the University of Ottawa. I asked them three questions. Who here feels like AI is making their degree less valuable? Nearly every hand went up. Who is worried that the job they imagined doesn’t exist anymore? Most hands stayed up. Who has no idea what the future looks like? Every single hand.
I want to be honest with you: your anxiety is not unfounded. This is a real issue. It’s beyond the scope of any one person — including me — to have a concrete answer. Even Geoffrey Hinton, one of the founding fathers of modern deep learning, has said he doesn’t know where this is headed.
But I think a lot of us — students, professionals, and organizations alike — are looking at AI through the wrong lens. And that framing mistake is causing a lot of unnecessary panic.
Stop Looking at Where AI Is. Look at the Slope.
The most common mistake I see — from students, from seasoned professionals, and even in government — is to take a snapshot of AI today, notice its limitations, and assume those limitations define a permanent boundary.
“It hallucinates.” “It can’t connect two systems reliably.” “It messed up the references.” All true. Today.
But four years ago, if I had told you that in 2026, AI would write fully functional applications from a single prompt, connect to live databases, and make engineers 80–90% more productive — you would have thought I was out of my mind. What AI could do four years ago was trivial compared to three years ago. Three years ago was trivial compared to two years ago. And so on.
When you look at the slope of that trajectory, not the current snapshot, the picture looks very different. And that slope is what companies are already planning around. Many organizations have quietly paused engineering headcount expansion — not because AI has automated everything today, but because they’re watching and waiting, hedging their bets on what they’ll actually need in five years.
This is the framing shift that matters: stop optimizing for what AI can’t do right now. Start building the capabilities that will matter no matter where the trajectory goes.
Five Pillars That Will Matter Regardless of Where AI Lands
I told the students there are five areas worth deliberately developing — not because they’re AI-proof, but because they compound in value as AI gets more capable, not less.
1. Judgment
AI can generate solutions. It can optimize for metrics. What it cannot do — at least not in any way people will trust — is make the judgment call about how a system should interact with human lives. Which stakeholders it affects. What trade-offs are acceptable given a specific community, context, and set of constraints.
That judgment is yours. And the more capable AI becomes, the more consequential your judgment over what to build, and what not to build, becomes.
2. Communication
The most underrated engineering skill has always been bridging the gap between technical and non-technical stakeholders. Most clients don’t know what they actually need — they know what they think they need. The ability to walk into a room, read the dynamics, earn trust, and translate between two completely different mental models of a problem: that is something AI will not replace.
Think about it: AI can teach you almost any subject. Any course, any concept, on demand. And yet most students still prefer a human professor with imperfect knowledge over a perfect AI tutor. That’s not irrational. That’s human. And clients feel the same way.
3. Learning to Learn
I hear this often from students: “Does it actually matter that I understand how a flip-flop works? AI already knows that.” It’s the wrong question.
Every concept you work through in an engineering program isn’t just about that concept. It’s about unlocking a region of your brain that knows how to reason about a new class of problem. When you work through the 25 basic micro-operations of a simple computer in a computer architecture course, you’re not memorizing trivia. You’re building the mental scaffolding that lets you pick up a new language, framework, or paradigm five years from now and get productive in days instead of months.
The specific knowledge you graduate with will almost certainly be obsolete. The ability to rapidly acquire new knowledge won’t be. That’s the actual deliverable of a rigorous engineering education.
4. Discipline and Depth
We live in an economy specifically designed to erode your ability to do deep work. To sit with a hard problem for two or three hours without switching tabs, without opening your phone, without skimming to “the good part.”
And ironically, AI is accelerating this erosion. People use AI to skip the thinking, not to augment it. As that becomes the norm, your ability to do genuinely deep, sustained, focused work becomes rarer — and therefore more valuable. Not as a personality quirk, but as a real professional differentiator.
5. Ownership and Ethics
AI is being deployed in healthcare, transportation, education, military systems, financial infrastructure. Everywhere. And when it fails — when the model is biased, when the edge case kills someone, when the output causes harm — someone is responsible. Someone signed off. Someone said: this is ready for production.
As AI handles more of the execution, the engineer’s role shifts toward ownership of the decision to deploy. The judgment of readiness. The accountability for downstream effects. That is not a diminished role. That is a more important one.
What This Means for How We Build AI Today
I work in both worlds — as an ML practitioner in industry and as a professor at the University of Ottawa. What I see consistently in both is that the companies getting the most from AI are not the ones that automate the most aggressively. They’re the ones that are the most intentional about where human judgment stays in the loop.
The honest assessment before you build anything: Is this the right problem to solve with AI? What does the failure mode look like? Who is affected if the model drifts? Does the slope of AI capability in this domain outpace your ability to oversee it?
These aren’t soft questions. They’re the questions that determine whether an AI investment pays off or becomes a liability. For a practical framework, see how to assess AI readiness before you spend anything. And they require exactly the kind of judgment, communication, and depth that no tool — however capable — can substitute for.
The value in the AI era isn’t in knowing how to write Python or build a model. It’s in knowing what to build, for whom, and whether you should build it at all.
Wassim El-Ahmar is an ML engineer and professor at the University of Ottawa, and the founder of Cedario Analytics. Cedario helps Canadian businesses cut through AI hype with honest feasibility assessments before committing to a build.
Leave a Reply