A number I keep coming back to this year: in the second quarter of 2026, Statistics Canada reported that 19.2% of Canadian businesses had used AI to produce goods or deliver services in the previous 12 months. Two years earlier, that figure was 6.1%.

That is a real shift. It is also a good example of why adoption statistics can be both important and incomplete. “Uses AI” can describe everything from a chatbot to a genuinely redesigned operating model.

Adoption is not transformation.

The OECD’s 2026 review of Canadian productivity treats AI as one possible source of productivity growth, but places it beside skills, investment, job mobility and broader structural change. That framing makes sense to me. Technology rarely arrives in a vacuum; its value depends on the work around it.

A firm can buy a capable tool and still leave the underlying process untouched. The result might be faster drafting, more output or lower effort in one part of the workflow without changing the quality of the decision or the overall economics.

The useful question is not “Are we using AI?” It is “Which work changed, which decisions improved, and what new failure modes did we create?”

The research is more interesting when it complicates the story.

The peer-reviewed “jagged technological frontier” study in Organization Science is a good example. On tasks inside the model’s capability frontier, participants completed more work, faster, at higher quality. On a task outside that frontier, AI access reduced the likelihood of reaching the correct answer.

That makes AI less like a universal productivity layer and more like a portfolio of task-level bets. Organizations still need to decide where automation is safe, where augmentation is valuable, where independent verification matters, and where a human should form a view before seeing the model’s output.

I would measure the workflow, not the excitement.

If I were evaluating an AI rollout, I would want a baseline before implementation and a small set of outcomes after it: cycle time, rework, quality, escalation rates, error severity, customer or employee experience, and whether capacity actually moved to higher-value work.

I would also ask who learned how to use the tool well. OECD work on AI and skills makes training sound less glamorous than models and agents, but it is probably one of the more practical determinants of whether adoption turns into value.

The story I am watching is not whether AI gets more capable. It will. The management story is whether organizations get better at redesigning work around it.

Sources + further reading

  1. Statistics Canada. Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026. June 11, 2026. Statistics Canada — AI use by businesses in Canada, Q2 2026 ↗
  2. OECD. Reviving Productivity Growth in Canada: The Role of Worker-Oriented Policies. June 29, 2026. OECD — Reviving Productivity Growth in Canada ↗
  3. Dell’Acqua F, et al. Navigating the Jagged Technological Frontier. Organization Science. 2026;37(2):403–423. Organization Science — Navigating the Jagged Technological Frontier ↗
  4. OECD. AI and skills: What we know so far. June 5, 2026. OECD — AI and Skills: What We Know So Far ↗

These are personal research notes and commentary. I link the underlying sources so the evidence can be checked directly.