CANAGI maps 516 Canadian occupations in one view. Each is drawn at the size of its 2023 workforce and coloured by one measure at a time: pay, projected demand, training pathway, or an experimental index of exposure to generative AI.
This guide shows how to read it when you are weighing a job, a program or a change of field, and what to check for yourself before you decide.
Read it as someone
Colour AI exposure index · Relative rank among 516 occupations, 0 to 10. Experimental CANAGI model.
Select a tile to pin its values here.
Summary
Every unit group in NOC 2021, Canada’s occupational classification
Those with a published 2023 employment count
In those occupations, 2023. Not job openings
AI exposure, projected outlook, pay and training pathway
A single interactive map of Canadian work, built from public Government of Canada data: 516 occupations in 10 career fields, from retail salespersons and visual merchandisers (580K people in 2023, the largest) to water well drillers (1,200, the smallest on the map).
Recolour the map by four measures, search and filter careers, open any occupation’s profile with its sources, and compare up to three careers or three career fields side by side.
Employment, pay, training, projected demand and AI exposure are usually published separately, on different dates and in different units. CANAGI puts them in one frame so you can see how they line up for the same job.
A forecast or a verdict. Pay is a published median, the outlook is a projection for 2024 to 2033, and the AI exposure index is an experimental ranking. None of them is a chance of losing a job.
02 Why I built it
Try answering a simple question about one job: is registered nursing a good field to go into?
The facts exist, and they are public. Statistics Canada and Employment and Social Development Canada (ESDC) publish how many people work in each occupation, what they typically earn, what training the work usually requires, and whether demand is projected to outrun supply. But each answer lives in its own release, and each describes a different moment.
For registered nurses, the employment count of 363,100 is for 2023. The median wage of $43.27 an hour describes 2023 to 2024, but it was released in November 2025. The training pathway, TEER 1, is a category from the 2021 classification. The outlook, a strong risk of shortage, is a projection for 2024 to 2033.
AI exposure is the least settled of all. Statistics Canada has published experimental research on how exposed occupations are to AI, and CANAGI builds its own experimental relative index from ESDC’s ratings of the activities each job involves. For nurses it is 6.3 out of 10: a rank, not a probability.
That is five answers from five releases, covering several reference periods and at least four kinds of unit: people, dollars an hour, categories and ranks.
Now compare nursing with two other careers you are weighing, and the bookkeeping triples before you have learned anything about which one suits you.
I built CANAGI to do that bookkeeping once, carefully, and to show the result in a single view. Every occupation sits on the same map. Every measure keeps its source, its date and its unit. Where a number is converted, such as an hourly wage made annual, the conversion is stated. Where a number is modelled, it says so.
What I did not want to build was a ranking. CANAGI does not score careers as good or bad, and it does not predict which jobs will exist in ten years. It shows how the published evidence lines up, so the person asking the question can do the judging.
03 How to read the visualizer
01 / 06
Each rectangle is a unit group from the National Occupational Classification, NOC 2021: the federal system that sorts every job in Canada by the work it involves. The tile picked out here is NOC 31301, registered nurses and registered psychiatric nurses.
There are 516 unit groups. The 485 with a published employment count appear on the map.
02 / 06
A tile’s area is proportional to the number of people employed in that occupation in 2023, the base year of ESDC’s Canadian Occupational Projection System (COPS).
Registered nurses: 363,100 people. Data scientists: 36,600, about one tenth of the area.
Area counts people who hold these jobs, not job openings. A large tile is a common occupation, not an easy one to enter.
03 / 06
Occupations are grouped by the first digit of their NOC code into 10 broad fields, outlined here. Sales & service is the largest, with 5M people. Senior management is the smallest, at 63,900.
These are occupation families, not industries. A hospital’s accountants sit in business and finance, not health.
04 / 06
CANAGI opens on the AI exposure index. Pick another layer and every tile is recoloured while the areas stay exactly where they are.
The colour scale runs from teal to rose on every layer, but its meaning changes. On pay, teal is higher pay. On AI exposure, rose is a higher relative score. On outlook, teal is a projected shortage. On training pathways, colour marks a category, not a better or worse career. Read the legend every time.
05 / 06
Colour is good for spotting clusters and poor for reading exact values, and not everyone sees it the same way. So CANAGI always writes the value down: on large tiles, in the tooltip when you hover or focus a tile, and in each occupation’s profile.
Try it. Hover or tap any tile, or focus the map and move with the arrow keys. Press Enter to pin a reading below the map.
06 / 06
31 occupations have no published 2023 employment count, so they cannot be sized. CANAGI keeps them in the Career Explorer list rather than guessing. Where any value is unavailable, CANAGI shows it as unavailable, in grey, and never as zero.
The map is national. It says nothing about a particular city or province, and nothing about the jobs being advertised today.
Colour AI exposure index
Hover, tap or arrow through the tiles. Select one to pin its values here.
That is the whole grammar. Area is the size of the workforce in 2023, fixed. Colour is one measure at a time, and its legend tells you which direction is which. Everything else in CANAGI, the filters, the profiles and the comparisons, is built on those two ideas.
04 From a question to a comparison
01 / 07
“I could train for a few different jobs. How do they compare?”
This is the most common kind of question CANAGI is built for. The steps that follow use a small, faithful slice of its Career Explorer: real records from its data release, and criteria that behave exactly as its filters do. Three careers are chosen to start. Change them in the next step.
02 / 07
These sixteen occupations were picked to span fields, pay levels and training pathways. CANAGI itself includes all 516.
3 of 3 chosen
03 / 07
CANAGI multiplies hourly medians by 2,080 hours so that every occupation sits on one annual scale. The result is a full time equivalent, not what any one person earns. Set a minimum to see which careers clear it.
04 / 07
The index places each occupation among 516 by how exposed its rated activities are to software based generative AI. A maximum here filters by that rank. It cannot filter out jobs at risk, because nothing in the index measures risk.
05 / 07
The outlook is ESDC’s projection of labour market balance for 2024 to 2033. The pathway is a training category from NOC 2021. Neither is a score: a projected shortage does not promise a job, and a higher TEER number is not a worse career.
Criteria combine with AND. Ranges include their endpoints. An occupation with no published wage cannot meet a pay minimum. These are CANAGI’s rules.
06 / 07
07 / 07
The comparison continues in CANAGI’s own side by side view, with the same careers and criteria. The address carries all of it, in the format CANAGI’s source code reads.
The link, part by part: https://www.canagi.me/1 CANAGI?compare=31301,72200,212322 Your careers, by NOC code&comparison=13 Opens the side by side view#careers4 In the Career Explorer
Annualized payCAD a year
A$90,002B$72,800C$100,006
AI exposure indexRelative, 0 to 10
A6.3 / 10B3.3 / 10C9.6 / 10
OutlookCOPS, 2024 to 2033
Strong surplusModerate surplusBalanceCModerate shortageBStrong shortageATraining pathwayNOC 2021 TEER
TEER 0TEER 1ACTEER 2BTEER 3TEER 4TEER 5Employed, 2023People, not openings
No criteria set yet, so all 516 occupations in CANAGI pass.
Nothing in the comparison ranks the careers. It lines up four published measures and one experimental one, each with its own date, so the trade offs are visible and the choice stays with the person asking.
05 The same product, different audiences
This is the same choice as the one at the top of the guide. It changes examples, notes and next steps throughout, is saved on this device, and travels in the page address, so a shared link opens the same version.
If you are weighing a career, CANAGI is most useful as a way to line up the questions you would ask about any job, and to spot the trade offs between a few of them.
Start with the careers you already have in mind. Read each measure as one piece of evidence about the national picture, not as a score for the job. The AI index, in particular, tells you where an occupation’s work activities rank in one model. It does not tell you whether the job will be there.
What matters first
Relative AI exposure index · bars show people employed in 2023
Worked example
NOC 31301
Registered nurses and registered psychiatric nurses
NOC 33102
Nurse aides, orderlies and patient service associates
Registered nurses and nurse aides share a field and a projected strong risk of shortage. Their AI index scores are far apart: 6.3 and 1.9.
Open both profiles and the same three activities lead each list of modelled contributions: documenting and recording information, processing information, and interpreting information for others. The difference is how highly OaSIS rates them. For nurses, those activities are rated 4 out of 5; for nurse aides, 2. Both are rated at or near the top for assisting and caring for others (4.8 and 5), which the model treats as hard for software to assist.
What you can take from this: OaSIS rates the information side of nursing as more demanding, and the model counts that as more room for software assistance. What you cannot: that either job is safer. The index does not model hiring, and both occupations are projected to need more workers.
06 What the numbers do and do not say
A number on a map looks equally solid whatever its origin. CANAGI tries to undo that impression by labelling where each value comes from and what was done to it.
The distinction that matters most is between what was observed, what was derived by arithmetic, and what was estimated with assumptions. A fourth kind, classification, covers categories and ratings assigned by a published standard.
| Measure | Kind | Source | Period | Coverage |
|---|---|---|---|---|
| Occupations and career fields | Classification | NOC 2021 version 1.0, Statistics Canada (opens in a new tab) | Classification adopted for 2021 | 516 of 516 |
| Employment (tile area) | Observed | COPS employment file, ESDC, based on the Labour Force Survey (opens in a new tab) | 2023 base year | 485 of 516 |
| Published median wage | Observed | Job Bank national wages, ESDC (opens in a new tab) | Released November 19, 2025. Reference: 483 records 2023 to 2024, 18 for 2024, 14 for 2021 | 515 of 516 |
| Annualized pay | Derived | CANAGI, from the published median (opens in a new tab) | Same as the wage it comes from | 515 of 516 |
| Outlook | Estimate | COPS projections, ESDC (opens in a new tab) | Projected 2024 to 2033; published January 30, 2025 | 485 of 516 |
| Training pathway (TEER) | Classification | NOC 2021, second digit of the code (opens in a new tab) | Classification adopted for 2021 | 516 of 516 |
| AI exposure index | Estimate | CANAGI model v1.0 using OaSIS 2025 activity ratings, ESDC (opens in a new tab) | OaSIS 2025 ratings; scored September 7, 2026 | 516 of 516 |
| Latest national employment (context only) | Observed | Statistics Canada, Table 14-10-0310-01 (opens in a new tab) | August 2026, seasonally adjusted | One national total |
Pay is the easier trace: one published figure, one stated conversion. The AI exposure index is the harder one, because it passes through assumptions. Try both, for any of the occupations, and notice where the kind of number changes.
ESDC’s Job Bank publishes wage estimates for every occupation, released November 19, 2025. These are the columns CANAGI reads for software developers and programmers, with the values it kept.
| Source column | Value |
|---|---|
| NOC_CNP | NOC_21232 |
| prov | NAT |
| ER_Code_Code_RE | ER00 |
| Low_Wage_Salaire_Minium | 30 |
| Median_Wage_Salaire_Median | 48.08 |
| High_Wage_Salaire_Maximal | 76.92 |
| Annual_Wage_Flag_Salaire_annuel | not 1, so hourly |
| Reference_Period | 2023-2024 |
| Revision_Date_Date_revision | 2025-11-19 |
| Data_Source_E | Labour Force Survey |
Download the full file from the Job Bank wages dataset (opens in a new tab). CANAGI’s release record lists its exact address and SHA-256 checksum.
The index answers one narrow question: compared with the rest of CANAGI’s 516 occupations, how much of this occupation’s rated work falls on activities that CANAGI’s rubric treats as open to assistance from software based generative AI?
Three features make it easy to misread. First, it is built on judgment: the 39 coefficients are CANAGI’s written assumptions, not measurements. Second, it is a rank. The raw scores behind it are bunched between 2.95 and 5.52, and ranking spreads them evenly across 0 to 10, which makes small differences look large (Fig. 6.3). Third, it is about activities, not jobs. Nothing in it describes whether employers adopt AI, how demand changes, or what happens to employment.
Relative AI exposure index, 0 to 10
The index is not
Open any of these for the rules behind the numbers. The main text does not depend on them.
For each OaSIS profile: raw score = 10 × Σ(activity level × capability coefficient) ÷ Σ(activity level), across all 39 activities. An all zero or incomplete profile stops the build instead of receiving an invented score.
When an occupation has several OaSIS profiles, their raw scores are averaged with equal weight, because employment weights for profiles are unavailable. The relative index is then 10 × (midrank − 1) ÷ 515 across the fixed cohort of 516 occupations, rounded to one decimal. Ties share a midrank.
Source: CANAGI exposure methodology (opens in a new tab) and model script (opens in a new tab).
Wages, employment, education, job titles, demographic characteristics and the COPS outlook do not affect an individual occupation’s score. Employment is used only later, as a weight in summaries such as the 5.2 employment weighted average.
Robotics is out of scope. Activities such as handling objects, operating vehicles and performing physical work are assigned a coefficient of 0.
Six scenarios combine linear or squared activity weights with coefficients lowered by 0.15, unchanged, or raised by 0.15, bounded to 0 and 1. Each scenario re ranks all occupations. The range shown is the lowest and highest resulting index.
It does not capture other plausible rubrics, uncertainty in the OaSIS ratings, variation within an occupation, real AI performance or future capabilities. A narrow range does not validate the model. 26 of the 106 occupations scoring 8 or higher fall below 8 in at least one scenario.
Each profile also shows a human oversight value: the higher of OaSIS’s ratings for consequence of error and for health and safety responsibility, on a 0 to 5 scale. It is displayed beside the index, never subtracted from it, and is not a validated adjustment.
Employment totals add distinct occupations with a known count (20,136,800 people across 485 occupations). This is below the COPS all occupations total because some employment is not available at this level of detail.
AI averages and distributions are weighted by 2023 employment and exclude occupations missing either value. Shortage share divides employment in moderate or strong shortage occupations by employment with an assessed outlook: 27%, or 5.4M people. Pay spreads treat each occupation’s median equally; they are not distributions of individual workers’ pay. Education mixes count occupations, not workers.
31 occupations have no published employment count at this level of detail, so they cannot be sized. The same 31 have no COPS outlook. They remain in the Career Explorer list. One occupation, senior managers in health, education, social and community services and membership organizations (NOC 00013), has no published national wage.
Newer employment tables exist, such as the monthly national total, but they use broader groups and cannot be divided into these occupations without assumptions CANAGI does not make.
07 The communication decisions behind the product
What this chapter can show
The reasoning behind each choice, and the reader’s task it was made for. Every screen below is CANAGI’s own interface, rendered from its source code.
What it cannot show
That any choice worked. No user research or usage data is on record for CANAGI, so none is claimed here. Each plate ends with a small test that could provide that evidence.
Where to look first Decisions 3 and 6 bear most on weighing careers: how pay is shown, and how the comparison stays out of your decision.
Plate 7.1CANAGI · Visualizer, layer controls
On the pathway layer the legend changes form: six numbered categories instead of a continuous scale. The TEER explanation on the summary cards below it adds that pink does not mean a career is worse.
Decision 1
Rationale only Not tested with users.
A test that would tell: Show readers the pay layer and ask what teal means. Compare wrong answers with and without the adjacent legend and caveat.
Plate 7.2CANAGI · Visualizer, summary figures
Decision 2
Rationale only Not tested with users.
A test that would tell: Ask readers what 20.1M counts. Count how many answer “job openings”.
Plate 7.3CANAGI · Career profile, headline figures
Decision 3 · If you are exploring careers
Rationale only Not tested with users.
A test that would tell: Ask readers to state the occupation’s published wage. Check whether they find the hourly figure and its period.
Plate 7.4CANAGI · Career profile, explaining the estimate
Decision 4
Rationale only Not tested with users.
A test that would tell: Ask “what does 8 out of 10 mean?” before and after readers see the label. Count answers that describe a percentage or a risk.
Plate 7.5CANAGI · Career profile, the evidence
Decision 5
Rationale only Not tested with users.
A test that would tell: Record how many readers open the calculation, and whether those who do not can still name one driver.
Plate 7.6CANAGI · Career comparison
Decision 6 · If you are exploring careers
Rationale only Not tested with users.
A test that would tell: Ask readers which career the comparison recommends. Count how many name one.
This guide made its own editorial choices: which example comes first, where a caveat goes, what to leave for later. Turn on Behind the explanation to see short notes about them in the margins throughout. They describe intent, like the plates above, not measured results.
08 Continue exploring
Next steps if you are exploring careers
Reading for someone else? Switch the reader at the top of the guide, in the chapter rail or in chapter five.
Official sources
Documentation
