How to think clearly when AI changes faster than you can read about it.

Why I focus on skills, not models

This is the core of how I lead AI transformation at Colliers.

Knowledge of a specific model loses value quickly. Becoming an expert in this quarter's tool is like learning every shortcut in an app just before a major redesign. The next release arrives, and much of that effort starts over.

Skills work the other way. They compound. Every improvement in AI makes a skilled person more valuable, because they can point more capability at better outcomes.

So instead of asking our teams to keep pace with every release, I invest in five skills that hold their value:

Problem framing. The most valuable AI skill is knowing what to ask. An engineer who can break a complex site assessment into clear, well-scoped questions will get better results from any model, today's or tomorrow's.

Judgment and verification. AI output is a draft, not a deliverable. In engineering, where safety and professional liability are real, the ability to review, challenge and validate output is non-negotiable. Our standards of care don't change when the tool does.

Domain depth. A model knows a little about everything. Our people know a great deal about real estate, buildings, infrastructure and markets. That depth is what turns generic output into client-ready work.

Communication and influence. As tools multiply, the scarce skill is helping people see why something matters and act on it. For our marketers, that's storytelling that moves an audience. For our engineers, it's explaining technical risk in language a client can trust.

Workflow thinking. Understanding how work moves from brief to delivery, and where it gets stuck, matters more than mastering any single tool. People who can redesign a process can plug in whatever model comes next.

Whether we're working with today's model or one five generations from now, none of these skills lose value. That's the point.

The 72-hour hype cycle

Watch any major AI release and you'll see the same pattern.

Hour one: the launch post goes live, with polished demos and bold claims.
By evening: the first "I tested it for 24 hours" videos are up.
Day two: benchmark charts, side-by-side comparisons, and threads crowning winners and losers.
Day three: hot takes on the hot takes.
By the end of the week: everyone has moved on to the next release.

Very few people go back to ask whether the last one changed anything real.

This isn't a conspiracy. It's the attention economy doing what it does. AI companies need launches to land. Influencers need fresh content. Platforms reward whoever is first and loudest. Everyone has a reason to make the moment feel bigger than it is.

What we end up with is a constant stream of first impressions and almost no long-term verdicts. For a business, first impressions are the least useful information there is. What I need to know is how a tool performs in month three, inside a real workflow, with real data, under real governance. Almost nobody publishes that.

So when something feels urgent, I ask one question:

Urgent for whom?

Usually it's urgent for whoever is selling it or talking about it, not for the team doing the work.

Utopia, apocalypse and the build-out

Alongside the launch hype runs a darker storyline. AI will take every job. AI will outsmart us. AI could end humanity. These headlines travel further than almost anything else in tech.

Some of these concerns are serious. Researchers, and many AI labs themselves, work on real safety risks, and that work deserves attention. I'm not dismissing it.

But context matters. Over roughly the past year, the industry has committed hundreds of billions of dollars to data centres, chips and power. At Colliers, we see this build-out up close. Data centres have become one of the most active corners of real estate.

An investment of that scale needs a story big enough to justify it. "A useful productivity tool" doesn't do that. "The most powerful technology in human history" does.

That's why the utopian promises and the apocalyptic warnings, which look like opposites, often have the same effect. Both make AI feel enormous, inevitable and urgent. And that feeling helps attract capital, talent and political attention.

I don't let either extreme set my agenda. Neither the miracle nor the catastrophe changes what my teams should do on Monday morning: pick a real problem, apply the right tool, check the output, and measure the result.

The 30-day rule

My teams follow a simple habit: if a release still matters in 30 days, we look at it properly.

Most don't. The excitement fades, the edge cases appear, and the gains turn out to be modest. The ones that survive have earned a real evaluation, and by then others have already found the rough edges for us.

The rule doesn't mean we ignore the frontier. A small group of us watches it closely, so the wider team doesn't have to. Our engineers and marketers are there to deliver for clients, not scroll through release notes.

Every use case has to give something back

Before any AI initiative gets resources, it has to answer one question: what does this give back?

There are only a few honest answers:

That last one is underrated. Sometimes the smartest AI decision is realising a process shouldn't exist at all. If a use case can't point to at least one of these, the model behind it doesn't matter.

Start with the problem, not the product

I push my teams to flip how they learn.

Don't start with a tool and look for uses. Start with a real problem. For example: "Preparing a feasibility summary takes too long. What's the fastest way to fix that?" Then go find the answer.

Learning that starts from a real need sticks. Learning that starts from a feed usually doesn't. And the best answer is rarely the newest one.