Math101 · Independent Ontario learning library
AI is already changing Canadian work. The evidence does not support either extreme claim that every professional job is disappearing or that nothing important will change.
The most defensible plan in July 2026 is to prepare for task redesign, uneven adoption, stronger verification demands, and a labour market influenced by far more than AI.
What current Canadian evidence says
Statistics Canada’s June 2026 worker-use study found that reported generative-AI use for work rose from 17% of workers in September 2024 to 30% in July 2025. Across the survey period, workers in professional, scientific and technical services reported particularly high use, followed by educational services and finance/insurance/real-estate-related industries. Workers with a bachelor’s degree or higher were much more likely to report use than workers with high school or less.
This establishes adoption, not replacement. A worker can use a tool while their occupation grows, shrinks, or changes for unrelated reasons.
Statistics Canada’s early employment analysis through December 2025 found overall employment continued to grow and found no clear evidence that jobs categorized as more exposed to and less complementary with AI had experienced disproportionate declines relative to other occupations. Employment growth was weaker for younger and less-educated workers across exposure groups, but the study cautions that demographic and economic forces overlap. It is early evidence, not proof that future effects will be small.
The correct summary is:
- AI use is rising quickly and unevenly.
- Exposure is not the same as job loss.
- Early Canadian employment data does not show a simple mass-displacement pattern.
- Entry-level tasks and pathways may still change before an occupation disappears.
- No credible source can identify a permanently “AI-proof” degree.
Exposure, complementarity, automation, and replacement are different
These terms are often collapsed online.
| Concept | Meaning | Example question |
|---|---|---|
| AI exposure | some occupational tasks overlap with capabilities of AI tools | Could a model draft, classify, predict, translate, or generate part of this work? |
| AI complementarity | AI may increase the worker’s productivity while human judgment remains important | Does the worker set goals, validate output, integrate context, and own consequences? |
| Automation exposure | equipment/software can execute a routine process with less human action | Can the workflow itself be mechanized or standardized? |
| Job replacement | fewer people are employed in the occupation because tasks, demand, organization, and labour supply change | Did employment actually decline for a defensible causal reason? |
Statistics Canada’s 2026 study of certified journeypersons emphasizes that AI and automation are different. Many trades have lower exposure to cognitive AI than office occupations while still facing some exposure to other forms of automation. Physical work is not magically technology-proof, and high AI exposure does not automatically mean high replacement risk.
Think in task bundles, not job titles
An occupation is a bundle of tasks performed inside an organization, market, legal system, and relationship network. AI may affect each part differently.
Tasks already becoming faster
- first drafts, summaries, and routine correspondence;
- search, document comparison, and information extraction;
- classification and coding of familiar cases;
- boilerplate code, tests, formulas, and simple analysis;
- transcription, translation, and meeting notes;
- image, layout, presentation, and content variations;
- first-pass customer or employee support;
- pattern detection and decision support in bounded systems.
Tasks whose importance may rise
- deciding what problem is actually worth solving;
- checking data provenance and whether a source exists;
- detecting plausible but wrong output;
- handling exceptions, ambiguity, and conflicting objectives;
- protecting confidential and personal information;
- integrating legal, physical, financial, ethical, and human context;
- negotiating, teaching, caring, persuading, and building trust;
- documenting accountability and explaining a decision;
- evaluating whether an AI system works across groups and failure cases.
AI can also create more work: review queues, cybersecurity incidents, privacy assessments, model monitoring, appeals, correction, and new service demand.
How major program families may change
Computing, data, and mathematics
AI can generate code and analysis quickly, which may reduce the value of producing routine output without understanding it. It increases the value of algorithms, systems, security, data engineering, evaluation, statistics, debugging, specifications, and domain context.
Students should still learn core programming without outsourcing every step. Build projects that show the problem, architecture, tests, failures, tool use, and independent reasoning. “I made five AI apps” is weak evidence if the student cannot explain or repair them.
Engineering and technology
AI may support simulation, optimization, design alternatives, predictive maintenance, documentation, and code. Physical constraints, measurement, standards, safety factors, field conditions, and professional responsibility remain. An attractive generated design is not evidence that a bridge, device, process, or control system is safe.
Accreditation and licensure do not disappear because a tool drafts work. The accountable practitioner still needs technical judgment.
Business, accounting, finance, and economics
Routine reporting, reconciliation support, research, spreadsheet creation, forecasting, and document review may become faster. Controls, audit evidence, tax/legal interpretation, model risk, client trust, and ownership of a recommendation remain. Junior roles may shift from producing a first draft toward reviewing, investigating exceptions, and improving systems.
Students should learn financial/accounting foundations, statistics, spreadsheets, data systems, privacy, and how to challenge a model. Avoid a curriculum that treats “AI” as a marketing elective detached from domain expertise.
Health and life sciences
AI may support imaging, triage, clinical documentation, research, scheduling, and decision support. Patient communication, consent, physical care, examination, uncertainty, professional standards, and responsibility remain. Health information is sensitive; unapproved tools can create privacy and safety risks.
Students need scientific reasoning, measurement, evidence appraisal, communication, and secure workflow habits. Tool fluency cannot substitute for an approved professional program or licence.
Education, law, policy, and social services
AI can draft lesson plans, summaries, forms, research notes, and standard communications. It cannot be assumed to understand a child, client, family, classroom, case record, legal duty, or local policy accurately. Relationship, safeguarding, advocacy, procedural fairness, and accountable professional judgment remain central.
Trades, construction, and physical operations
AI may affect estimating, scheduling, diagnostics, documentation, procurement, and predictive maintenance. Robotics and automation can affect physical processes separately. Site variation, dexterity, safety, troubleshooting, codes, and coordination remain. Students should add digital drawings, controls, diagnostics, and business skills to a strong trade foundation where relevant.
Creative and communications work
AI can generate abundant drafts, images, audio, and variations. This pressures generic production and may increase the importance of direction, research, taste, rights clearance, brand context, audience knowledge, editing, and original access. A portfolio should reveal the creator’s decisions and process, not only polished output.
What not to choose on faith
“Prompt engineer” as a whole career plan
Prompting is useful, but interface techniques change. A durable role usually adds domain knowledge, workflow design, evaluation, data, security, user research, or technical implementation. Treat prompting as a literacy layer, not a guaranteed stand-alone occupation.
A new degree because it has “AI” in the title
Inspect required calculus, linear algebra, probability, statistics, programming, algorithms, data systems, ethics, and capstone work. Ask when students enter the specialization and whether progression is competitive. A rigorous CS, mathematics, statistics, engineering, or domain degree with strong AI options may be better than a shallow title; the reverse can also be true. The calendar decides.
A career advertised as impossible to automate
Physical presence, trust, licensure, or human relationships can reduce some substitution risk, but tasks, staffing models, and technology still change. Choose work you can adapt within, not a supposedly frozen job description.
The five-layer capability stack
1. Domain knowledge
Know enough accounting, biology, engineering, law, logistics, education, or another field to detect nonsense and understand consequences.
2. Quantitative and data literacy
Understand rates, uncertainty, samples, correlation versus causation, measurement error, spreadsheets, and the data lifecycle. Add statistics or coding to the depth your field requires.
3. AI and systems literacy
Understand what the tool is being asked to do, what data reaches it, likely failure modes, evaluation, security, bias, and integration into a workflow. Tool-name familiarity is not systems literacy.
4. Human and organizational skill
Write, listen, ask precise questions, negotiate trade-offs, teach, collaborate, and make work usable by someone else.
5. Responsibility
Know when not to use the tool. Preserve confidentiality, document sources and decisions, disclose material tool use, escalate uncertainty, and own the final result.
These layers reinforce one another. “Soft skills only” and “technical skills only” are both incomplete.
A twelve-week student plan
Weeks 1–2: establish a baseline
Complete a small field-relevant task without AI. Save the work, time taken, errors, and explanation. This establishes what you can independently do.
Weeks 3–4: learn verification
Ask an AI system to solve or critique comparable tasks. Check every factual source, calculation, assumption, citation, and edge case. Keep a failure log.
Weeks 5–6: learn privacy and policy
Read the school or employer’s current rules. Read the tool’s data controls. Practise replacing personal/confidential data with safe synthetic examples. The Canadian privacy regulators’ generative-AI principles emphasize lawful authority, appropriate purpose, necessity, openness, accountability, safeguards, and accuracy.
Weeks 7–8: improve a workflow
Use AI for a bounded step—brainstorming test cases, explaining an error, comparing two public documents, or drafting a structure. Measure time and quality against the baseline. If it is worse, say so.
Weeks 9–10: add domain depth
Choose one result that required real subject knowledge to validate. Explain why a non-expert might accept the wrong output.
Weeks 11–12: publish an honest case study
Show:
- problem and constraints;
- your independent contribution;
- tools and data used;
- prompts or workflow where disclosure is allowed;
- evaluation method;
- errors found and corrections;
- privacy/copyright considerations;
- what you would not automate;
- final result and remaining limits.
This is stronger evidence than presenting generated polish as personal mastery.
A program-comparison scorecard for an AI-shaped market
Score exact programs from 1 to 5 and choose personal weights from 0 to 5.
| Axis | What to verify | Weight | Score |
|---|---|---|---|
| Foundation depth | core theory/methods that remain useful when tools change | ||
| Practice | co-op, labs, placements, research, capstone, real constraints | ||
| Evaluation | statistics, testing, evidence appraisal, safety, audit, model limits | ||
| Domain depth | enough subject knowledge to judge consequential output | ||
| Systems/data | databases, workflows, privacy, security, interoperability | ||
| Human responsibility | writing, teamwork, ethics, client/patient/user context | ||
| Career gate | accreditation, licensure, certification, graduate-school risk | ||
| Cost and flexibility | net cost, ability to pivot, electives, extra-year risk |
This is Math101 judgment, not a forecast. A program with no “AI” label can score extremely well.
Privacy and academic integrity
Never assume a consumer AI service is private. Do not enter identifying student records, medical information, client data, unpublished research, workplace documents, passwords, private source code, or assessment materials without authorization and an approved tool/process.
Academic-integrity rules are institution-, course-, and assessment-specific. “AI is allowed at my university” is not enough. Ask what assistance is permitted, what must be disclosed, how citations should work, and what independent competence must be demonstrated. Preserve drafts and sources.
The safest learning pattern is:
- attempt;
- ask for a hint, critique, or test—not necessarily a finished submission;
- verify independently;
- redo the core work without the tool;
- disclose use as required.
How parents can talk about AI without panic
Replace “Will this degree exist?” with:
- Which tasks are changing now?
- Which foundational knowledge lets a worker verify the tool?
- What experience will prove the student can work without and with AI?
- What professional or privacy duties constrain use?
- What alternate roles share the same capability stack?
- What evidence would change our view next year?
Students do not need to predict 2040 correctly. They need a strong next platform and a habit of updating.
Which degree is safest from AI?
No degree is guaranteed safe. Prefer deep foundations, real practice, adaptable methods, and work where you can combine tools with domain judgment, relationships, physical or system context, and responsibility.
Should every student learn to code?
Everyone benefits from computational and data literacy, but coding depth should fit the route. A future software engineer needs much more than a future nurse or teacher; all three need to understand what automated systems can get wrong.
Will AI eliminate entry-level jobs?
It may reduce or redesign some junior tasks, but current Canadian evidence does not establish a simple economy-wide elimination pattern. Students should build supervised experience, verification ability, and ownership of a complete workflow rather than relying on routine production alone.
Is AI use on a résumé dishonest?
Not when it is accurate and permitted. State the problem, your work, the tool’s role, and how you evaluated the result. Never claim generated output as a capability you cannot reproduce or defend.
Research and policy sources
- Statistics Canada: workplace AI use — worker-reported adoption from September 2024 to July 2025.
- Statistics Canada: early Canadian employment trends in the generative-AI era — employment patterns through December 2025 and important causal limits.
- Statistics Canada: AI and automation exposure among certified journeypersons — the distinction between cognitive AI exposure and automation.
- Office of the Privacy Commissioner of Canada: responsible, trustworthy, privacy-protective generative AI — Canadian privacy principles and accountability.
- Government of Canada: Learning together for responsible AI — public AI-literacy framework.
- Canada Job Bank trend analysis — occupation-specific wages, outlooks, and requirements rather than AI speculation alone.
Evidence boundary: adoption and employment statements above are tied to the cited Canadian studies. Career-family interpretations, capability stack, twelve-week plan, and scorecard are Math101 guidance. AI systems, institutional rules, labour markets, and law change; recheck current policy before consequential use.
