Democratize AI or risk a widening productivity divide

You can read the Globe and Mail piece and walk away with one headline-grabbing line: AI superusers are pulling ahead and widening the productivity gap. That’s not wrong; it’s just incomplete. The more consequential split isn’t simply between clever individuals and everyone else — it’s between organizations that are quietly rewiring how work happens and those still treating AI like a shiny browser tab.

Let’s start with where the article is absolutely right: some people are wringing far more value out of these tools than others. Yeah, no, that’s not unique to AI. Early web devs who really understood JavaScript looked like sorcerers. Excel power users basically ran half of corporate finance for a decade. Any tool that rewards practice and experimentation will produce “superusers.”

But stopping the analysis at “some people are better at this” misses the bigger choke point: data and workflow integration. AI in a vacuum is mostly parlor tricks. The real gains come when systems, permissions and information flows are redesigned so the tools can actually do something beyond spitting out a clever paragraph.

Take two sales reps. One lives inside a unified CRM, has clean customer histories, standardized messaging templates and explicit permission to experiment with prompts and automation. The other copies queries into a free chatbot and manually pastes answers into emails. Same underlying model, wildly different outcomes. The “superpower” isn’t the human; it’s the plumbing and the latitude to use it.

You see the same pattern in product teams. A manager with instrumented telemetry, internal search over support tickets and a lightweight internal model can iterate features based on real feedback. Another manager is guessing from anecdotal complaints and a quarterly survey. One looks like a genius; the other looks slow. What’s really different is the infrastructure and the habit of routing decisions through it.

Once AI is stitched into that fabric, network effects kick in. Faster experiments lead to better results, which free up more budget and trust, which pay for better tools and more data work, which make the next round even faster. It’s the same flywheel that helped Google and Amazon turn early infrastructure bets into durable advantages: scale your systems, then your systems scale you. Isaac Asimov played with this kind of institutional momentum in the Foundation series — once a big system starts compounding, individual heroes matter a lot less than the machinery they’re sitting on.

This is why I get twitchy when the conversation fixates on “teach your staff to prompt.” Sure, but prompt tricks are the shallow end of the pool. Training that actually shifts the curve looks more like: here’s how to redesign your intake process so AI can draft the first pass; here’s how to set guardrails; here’s how to monitor outcomes and feed that back in. That’s product thinking and change management, not just clever phrasing.

Companies like GitHub with Copilot, or Microsoft bundling assistants into Office, are betting that embedding AI directly into workflows beats making people tab over to a separate app. They’re not just handing out smarter UIs; they’re trying to sit where the work already lives. The Globe and Mail angle on “superusers” catches the people who adopt this early, but the deeper story is which orgs are willing to restructure so those people don’t drown in bureaucracy.

Policy folks have skin in this game too. If the productivity gains cluster only in firms that already have clean data and modern stacks, you get a reinforcing loop: rich in-house infrastructure makes AI wildly effective, which widens the gap with everyone still on legacy systems and paper approvals. That’s where ideas like adult reskilling, standards for data portability inside firms and targeted support for small-business modernization stop being buzzword salad and start being economic strategy. Without that scaffolding, the “superuser premium” can harden into structural advantage for early adopters.

There’s a popular counter-argument that says, relax, interfaces will democratize all this. As tools become easier and assistants get baked into email, docs and chat, the gap will flatten because you won’t need power skills. Funny thing is, that logic has been tried. Low-code and no-code platforms did widen access, but the biggest returns still went to the teams that changed their processes to match the tools, not just clicked around the new UI.

User-friendly assistants will absolutely lift the floor. The ceiling, though, is set by how much an organization is willing to re-architect: customizing models (or at least workflows) to domain-specific data, building feedback loops into daily work, rethinking incentives so people aren’t punished for trying to automate part of their role. Those are political projects as much as technical ones.

So if you manage people, don’t start with “who are my superusers?” Start with a map. Where does knowledge actually sit? Which roles guard unique data? Where do decisions get stuck waiting for approvals that could be automated or at least prepared by a model? Clear one or two of those logjams and you’ll often see outsized impact without touching the underlying AI.

Then, buy your team slack time to tinker. You can’t schedule breakthroughs in 30-minute slots between status meetings. Let people run small experiments on low-risk tasks, share what works and fold that back into the official process. And please stop grading productivity purely by visible effort or hours logged — AI changes the shape of work, not just its speed, and old metrics will treat the biggest wins as suspicious laziness.

The Globe and Mail piece is right that AI superusers are pulling ahead; they are. But the next headline won’t be about a handful of prompt wizards — it’ll be about the companies that quietly rebuilt their org charts and data stacks so those wizards became the norm instead of the exception.