of Canadian businesses reported using AI in Q2 2026.
That is more than triple the 6.1% reported in 2024.
Statistics Canada A field guide to adoption capacity
AI adoption is not
a tool decision.
It is a change in how people, organizations and communities decide, learn and work.
Read the argument Evidence-led · Canada · 2026 · 18 min
Interest is widespread.
Working systems are not.
The decisive question is not whether an AI system can produce an impressive result. It is whether people can use it reliably, responsibly and repeatedly in the conditions where real work happens.
AI adoption is often treated as a sequence of tool choices: select a platform, train users, count licences. That sequence mistakes access for capability. A useful demonstration can collapse when it meets sensitive information, ambiguous judgement, fragmented workflows, unequal access, weak oversight or a team with no time to learn.
New Manual proposes a different unit of progress: adoption capacity—the practical ability to identify worthwhile opportunities, test them safely, redesign the surrounding work, govern their use, develop people’s judgement and learn from evidence over time.
What the evidence says
Use is rising.
Results depend on the work.
Three signals describe the current moment: use is spreading, outcomes vary sharply by task and context, and organizational returns lag behind individual experimentation.
of Canadian workers had used generative AI for work.
Most users applied it to some—but not most—of their tasks.
Statistics Canadaof rural businesses reported AI use, versus 21.0% of urban businesses.
Adoption capacity is also an access and ecosystem question.
01Use is rising faster than operating capability.
+
In Canada, business AI use reached 19.2% in the second quarter of 2026—more than triple the 2024 rate. Yet adoption remains uneven: 21.0% of urban businesses reported use compared with 9.9% of rural businesses.
Read the source02Task performance can improve—and still mislead.
+
Field and experimental studies find substantial gains on some tasks, especially for less-experienced workers. Other studies show performance can deteriorate when a task falls outside the model’s uneven capability frontier.
Read the source03The unit of adoption is the workflow.
+
A tool login says little about whether a practice is useful, repeatable, governed or sustained. Research increasingly distinguishes firm, function and task-level adoption—and top-down provision from bottom-up use.
Read the source04Human judgement is a designed control.
+
Human review works only when reviewers have time, competence, authority and a clear standard. Overreliance is not solved by writing “human in the loop” into a policy.
Read the source05Participation and learning are infrastructure.
+
People need more than awareness: practice, feedback, peer support and the ability to challenge a proposed system. Social dialogue matters when roles, skills and job quality may change.
Read the source06Scale changes the problem.
+
A personal practice, an organizational system and a community program require different evidence, rights, resources and forms of stewardship. Readiness is therefore contextual—not a universal score.
Read the sourceThe adoption gap
Between a promising output and a working system sit the conditions most adoption programs leave implicit.
Access
→A tool is available.
Use
→Someone tries it.
Practice
→A task changes.
Capacity
The system can learn.
The proposed method
Meet the
MANUAL Method.
A six-move cycle for turning AI interest into responsible working capability.
It integrates technology-adoption research, implementation science, human factors, responsible-AI guidance and place-based capacity building. It is an applied method—not a certification or universal maturity score.
01
M
Map
the work and outcomes
Guiding question
What work matters, to whom, and what would better look like?
Minimum output
A task map and a small set of observable outcomes.
The stop rule: “No AI” is a successful decision when value is weak, safer alternatives are better, or risk cannot be brought within an acceptable boundary.
The assessment architecture
Six conditions.
Three nested scales.
Readiness is not one number. The Capacity Map makes the enabling conditions visible, then asks where the change must hold: in a person’s practice, an organization’s operating system, or a community’s support ecosystem.
Community ecosystem
Organization system
Person practice
01
Purpose & valueClarity before capability
02
Task & decision fitThe work, not the demo
03
People, agency & inclusionAdoption is social
04
Data, tools & infrastructureUsable foundations
05
Accountability & assuranceResponsibility stays human
06
Learning, evidence & sustainmentEvidence before expansion
Two-minute reflection
Where is your adoption capacity today?
Move each marker to the level best supported by evidence—not aspiration. This reflection is stored nowhere and is not a validated diagnostic. Its purpose is to improve the next conversation.
From assessment to decision
Do not pilot
everything promising.
Each use case is judged on value, risk and readiness. The route follows the evidence; an overriding risk gate can stop a use case regardless of its score.
Pilot
Run a bounded real-work test with a baseline, review points and a stop condition.
Prepare first
Fix data, policy, role, process or learning conditions before testing.
Park
Record the opportunity and the condition that would justify reconsidering it.
Simplify or skip
Improve the process without AI, choose a safer tool, or stop.
Three calls to action
Choose the scale
where change must hold.
The MANUAL cycle stays consistent. The stakeholders, evidence and support system change with the scale.
People
Build capability around work you actually do.
Individual adoption is not a prompt collection. It is the ability to choose appropriate tasks, protect information, verify outputs and remain accountable for the result.
Start here
- Choose one recurring, low-consequence task.
- Write down how you do it now and what quality means.
- Test AI on examples you understand well.
- Compare time, quality and effort—not novelty.
- Keep, revise or reject the workflow.
A useful first win is explainable, repeatable and leaves you more—not less—able to judge the work.Start a personal practice review
The evidence contract
Agree what proof means before the pilot.
A credible adoption decision considers six dimensions together. Speed alone is not success.
Outcome value
Did the work improve in a way that matters?
Quality
Was the result accurate, useful and fit for purpose?
Adoption
Was the workflow accepted, feasible and repeatable?
Human impact
What changed in agency, skill, workload and inclusion?
Risk & assurance
Did controls work and were incidents understood?
Sustainability
Can the practice be supported, governed and improved?
Operating principles
Understand the context before choosing a tool.Build with the people affected.Test before expanding.Prefer evidence over performance.Keep responsibility human.Be honest when AI is not the answer.
Do not begin with
“Which AI should we buy?”
Start with an honest picture of the organization, the people involved and the outcomes that matter. Identify where AI may fit, what responsible use would require and the smallest useful step that can resolve the next question.
New Manual
Build the adoption program around the people and work that matter.
Begin the assessment Download the full white paper For organizations, trusted networks and public learning programs across Canada.
Research base and selected sources+
Adoption and implementation. Venkatesh et al., UTAUT; Tornatzky & Fleischer, TOE; Damschroder et al., updated CFIR; Greenhalgh et al., NASSS; Proctor et al., implementation outcomes; RE-AIM.
AI at work. Brynjolfsson, Li & Raymond; Noy & Zhang; Dell’Acqua et al.; Bonney et al.; Stanford SIEPR Firm Data on AI; Statistics Canada business and worker surveys.
Human factors and learning. Long & Magerko on AI literacy; Lee et al. on critical thinking; Buçinca et al. on cognitive forcing; UNESCO AI competency frameworks; OECD and ILO skills and social-dialogue research.
Governance and place. NIST AI RMF Generative AI Profile; ISO/IEC 42001; Canadian privacy commissioners; Government of British Columbia and Government of Canada guidance; OECD and G7 SME adoption work; FNIGC OCAP principles.
Method note. The MANUAL Method is New Manual’s applied synthesis of the research traditions and guidance above. It has not been represented as a psychometrically validated instrument. Scores are conversation prompts, not rankings or certifications. This field guide is educational and does not replace legal, privacy, security, labour-relations or professional advice. Research current to 24 August 2026.