What DeepSeek Made Me Question About How We Lead in Canada

I went looking for a technology story. DeepSeek handed me a leadership one instead, and it wasn't flattering.

Here's the detail that stuck with me. GPU capacity is one of the most expensive things an AI company owns. At most organizations, getting access to it means a business case, a budget line and a stack of managerial signatures. At DeepSeek, a researcher with a promising idea can reportedly just use it. No approval queue.

They made that call on purpose. They decided the cost of slowing down their best people was bigger than the cost of occasionally wasting computer.

Read that again, then think about your own sign-off process for something far cheaper than a cluster of GPUs.

The rest of the picture is just as unusual. DeepSeek reportedly built a model that competes with the biggest labs in the world using a team of roughly 150 people, few management layers and no traditional individual KPIs. The whole thing runs on one bet: hire capable people, give them the tools, then trust them to use their judgment.

I can't stop thinking about that bet, because it runs against almost everything we've normalized here.

Are we managing risk, or managing our own anxiety?

Canadian organizations love controls. Approvals, reporting requirements, dashboards, committees, policies, layers. Some of that is necessary. Most of it started with good intentions

But be honest for a second. How much of it genuinely improves the work, and how much of it just makes us feel more comfortable?

I ask because I've been the problem. When I first moved into sales management, my instinct was to manage activity. Calls made. Meetings booked. Proposals sent. CRM fields filled in. Activity felt measurable, and measurable felt controllable, and controllable felt safe.

I still feel the pull. When results get uncertain and the pressure is on, my grip tightens like everyone else's.

Here's what I've learned it costs. The harder you manage every action, the faster your most resourceful people check out. The control you added to lift performance quietly starts lowering it. People stop experimenting. They stop challenging the process. They stop deciding anything.

They just wait for permission.

Canada's engagement problem is a productivity problem

Only 21% of Canadian employees are engaged at work, according to Gallup's 2026 State of the Global Workplace report.

That isn't a soft HR number. That's an enormous amount of talent, creativity and judgment sitting idle inside Canadian companies.

At the same time, the Bank of Canada has called our weak productivity an emergency. In 1984, we produced about 88% of the value the U.S. economy generated per hour worked. By 2022, we'd fallen to 71%, with only Italy sliding further among our G7 peers.

Engagement isn't the whole story. We also have weak business investment, thin competition and slow technology adoption. But we should at least be willing to ask whether the way we manage people is part of the drag.

We have one of the most educated populations on earth. We attract capable people from everywhere. We have strong industries and real resources.

Maybe we don't have a talent shortage.

Maybe we've just gotten exceptionally good at containing it.

We built the tech stack and forgot the human stack

For years, CRM vendors, and now the AI companies, have sold us the same seductive idea: track every call, keystroke and pipeline stage, and you can engineer predictable performance.

But Jim Collins had this right in Good to Great. Technology accelerates momentum, it doesn't create it. It can make a good team better. It can't rescue a broken culture.

We got so busy building the tech stack that we forgot about the human stack. You still need the right people on the bus, in the right seats. Drop talented people into a system that micromanages them into data-entry robots and you shouldn't be shocked when the engagement drains out of them.

The best sales teams I ever built still had accountability. We measured results. We inspected the pipeline. We used the CRM.

But my one-on-ones weren't KPI interrogations. They were support sessions. What are you trying to accomplish? Where are you stuck? What do you need from me? What can I get out of your way?

The data fed the conversation. It didn't replace it. A CRM is supposed to give a leader visibility, not become an electronic leash.

It's the same uncomfortable step as parenting

I'm relearning all of this at home, too.

Watching my sons move into young adulthood, the parallel to leadership is hard to miss. You give them your values. You hand them the tools. You offer advice and support.

And then you have to let go.

Is it stressful? Completely. Is there risk? Of course there is. But you can't helicopter a young adult into a capable life any more than you can micromanage an employee into an independent problem-solver. Past a certain point, the control works against the very thing you're trying to build.

Trust always carries risk. That's what makes it trust.

DeepSeek isn't a model we can copy literally

Before anyone emails me: no, a Canadian manufacturer, hospital, energy provider or construction firm can't run like an AI research lab.

Safety standards, engineering requirements, customer commitments and regulation all demand structure. Nobody wants freewheeling experimentation on an electrical install or a lockout protocol. Autonomy is not the same as abandoning accountability.

And DeepSeek knows it. By its own account, the company tightens into a coordinated, top-down structure when a major release needs specific people executing specific jobs.

So the lesson isn't "get rid of management." It's to be far more deliberate about where control is genuinely required. Govern the real risks tightly. Give people room everywhere else.

That might mean letting a frontline employee solve a customer's problem without a manager's sign-off. Giving a salesperson latitude on how they work a territory while still owning the result. Letting a plant employee test a small improvement without clearing three committees. Asking managers to remove barriers instead of monitoring every movement. And accepting that capable people will sometimes make a call differently than you would.

The questions worth sitting with

DeepSeek is really just a prompt to look hard at what we've normalized:

How many approvals does it take to act on a good idea here?

How much time do people spend reporting work instead of doing it?

Do our KPIs help people make better decisions, or help managers feel in control?

Are our managers clearing obstacles, or creating them?

Do the people closest to the customer actually have the authority to fix the customer's problem?

Are we hiring people for their judgment and then quietly refusing to let them use it?

Are our best people getting more meaningful work, or just more work?

None of those are soft questions. They're productivity questions.

We won't fix Canada's productivity decline by asking people to work longer, buying another piece of software, or measuring every action more aggressively. Technology matters. Investment matters. Skills matter. Accountability matters. But so do trust, autonomy and the speed at which good people are actually allowed to move.

The most uncomfortable thing about DeepSeek isn't that its approach is impossible for us. It's that parts of it are entirely possible... and most of us would still struggle to hand over that much control.

Maybe the real obstacle was never our people's willingness to perform.

Maybe it's our willingness, as leaders, to let them.

Canada has the talent. The only open question is whether our organizations are built to unleash it, or to contain it.

In short

DeepSeek reportedly built a frontier AI model with a team of about 150 people, no traditional KPIs, few management layers, and researchers who can access expensive computing power without a lengthy approval process. That model is a hard mirror for Canadian leaders. Only 21% of Canadian employees are engaged at work (Gallup, 2026), and the Bank of Canada has called weak productivity an emergency, noting Canada fell from 88% of U.S. output per hour worked in 1984 to 71% by 2022. The takeaway isn't to abandon accountability or copy an AI lab. It's to govern genuine risks tightly and give capable people freedom everywhere else.

Key takeaways

• DeepSeek reportedly built a competitive frontier model with about 150 people, no traditional individual KPIs, a flat structure, and researcher access to expensive GPU compute without a lengthy approval process.

• Canadian engagement is low: only 21% of employees are engaged at work (Gallup, 2026), meaning a large amount of talent and judgment is sitting dormant.

• Canada faces a productivity emergency: the Bank of Canada notes productivity fell from 88% of U.S. output per hour worked in 1984 to 71% by 2022, with only Italy declining more among G7 peers.

• Control is not free. Over-managing activity pushes capable people to disengage, so the more deliberate move is to govern real risks tightly and grant autonomy elsewhere.

• Technology is an accelerator, not a cure. As Jim Collins argued in Good to Great, it speeds up momentum a team already has; it cannot rescue a broken culture.

Frequently asked questions

What is DeepSeek's management model?

Founder Liang Wenfeng has described an organization with few conventional management layers, no traditional individual KPIs, and considerable freedom for researchers to pursue promising ideas, including access to expensive computing power without a lengthy approval process. These details come largely from Liang's rare interviews and secondary reporting, so treat them as company-reported rather than independently audited.

Does removing KPIs and controls mean removing accountability?

No. Autonomy is not the absence of accountability. Well-run teams still measure results, inspect the pipeline and use their tools. The distinction is between measuring outcomes and governing genuine risk versus monitoring every action. DeepSeek itself reportedly shifts into a coordinated, top-down structure when a major release requires specific execution.

Why does employee engagement matter for productivity in Canada?

Only 21% of Canadian employees are engaged at work (Gallup, 2026), which means a large share of available talent, creativity and judgment is sitting idle. Engagement isn't the only driver of productivity, but with the Bank of Canada calling weak productivity an emergency, it's worth asking whether management systems that contain people are part of the problem.

Can Canadian companies actually copy DeepSeek?

Not literally. A manufacturer, hospital, energy provider or construction firm operates under safety, engineering, customer and regulatory obligations that demand structure. Nobody wants freewheeling experimentation on a lockout protocol or a medical procedure. The transferable lesson is to be far more deliberate about where control is genuinely required, and to loosen it everywhere else.

Isn't more measurement and technology the answer to low productivity?

Technology, investment and skills all matter, but technology is an accelerator rather than a cure. As Jim Collins argued in Good to Great, it speeds up momentum a team already has; it can't fix a broken culture. Tracking every keystroke can even backfire by pushing capable people to disengage.

What questions should Canadian leaders ask about their own controls?

A useful starting set: How many approvals does it take to act on a good idea? How much time do people spend reporting work rather than doing it? Do our KPIs help people make better decisions or just help managers feel in control? Do employees closest to the customer have the authority to solve the customer's problem? Are we hiring people for their judgment and then preventing them from using it?

References

Collins, J. (2001). Good to great: Why some companies make the leap... and others don't. HarperBusiness

Gallup. (2026). State of the global workplace: 2026 report. Gallup, Inc.

Liao, R. (2025, January 30). Culture matters at DeepSeek. Leapfrog. https://leapfrogs.substack.com/p/culture-matters-at-deepseek

Rogers, C. (2024, March 26). Time to break the glass: Fixing Canada's productivity problem [Speech]. Bank of Canada. https://www.bankofcanada.ca/2024/03/time-to-break-the-glass-fixing-canadas-productivity-problem/

Schneider, J. (2025, January 30). DeepSeek's secret to success. ChinaTalk. https://www.chinatalk.media/p/deepseeks-secret-to-success