TL;DR. Most career advice is about playing harder. Strategy is about choosing the game. In May 2025 US wage data, the 90th-percentile worker earns 4.12 times the 10th-percentile worker across the whole economy, but only 2.29 times inside the typical occupation: much of the pay gap sits between arenas, not within them. A CEOtudent join of Bureau of Labor Statistics (BLS) AI exposure categories with OEWS pay percentiles for 830 occupations shows that arenas rated “Very high” for AI exposure have the highest typical pay (median of occupation medians $78,105) but the slowest projected growth (2.0 percent for 2025-2035), the fewest openings per 100 workers (8.4 a year) and the widest pay spread (median P90/P10 of 2.70). The CEO move: choose the arena where your relative edge is largest and the payoff shape fits the rank you can realistically reach, then write down what you will not compete on. The student move: learn the one specific edge that arena rewards, not a general list of trending skills.
This piece is about arena selection: where to play. It sits next to the published pieces on what career capital compounds and what decays, which covers the assets you build, and on how to build a personal thesis, which covers the point of view you hold. The question here comes first: which game are you entering, and why that one?
Why is choosing an arena a strategy question, not a planning question?
In “What Is Strategy?” (Harvard Business Review, November-December 1996), Michael Porter separates operational effectiveness, doing similar activities better than rivals, from strategy: “Strategy is the creation of a unique and valuable position, involving a different set of activities.” And, in the line that matters most for a career, “The essence of strategy is choosing what not to do.”
Porter also names what happens when everyone improves the same way, competitive convergence: “The more benchmarking companies do, the more they look alike.” The individual version is easy to see in 2026. When every analyst, writer and consultant uses the same AI assistants and the same courses, getting faster at the same work is operational effectiveness. It is necessary, and it is not a position. The published piece on building authority when everyone has the same AI tools covers the visibility side of that problem; this piece covers the choice that precedes it.
Roger Martin put it more plainly in “The Big Lie of Strategic Planning” (Harvard Business Review, January-February 2014): “Characterizing the key choices as where to play and how to win keeps the discussion grounded.” He also warns about companies that “prefer a small slice of a huge market to a large slice of a small one.” Translated to one person: “I work in tech” is not a where-to-play choice. It is a description of a crowd.
For an individual, the two choices become:
- Where to play: a specific occupation, client type or problem class where you compete for income and opportunities.
- How to win: the edge that makes someone choose you over the next qualified person in that arena.
The CEO + Student lens maps onto these directly. The CEO owns the first choice and the list of things you refuse to compete on. The student owns the second: building the edge that this arena, not a generic labour market, rewards.
What does comparative advantage mean for one person?
The oldest strategy logic for individuals is a footnote in David Ricardo’s On the Principles of Political Economy and Taxation (first edition 1817, third edition 1821). In chapter 7, Ricardo describes two men who both make shoes and hats, one better at both. He can “only exceed his competitor by one-fifth or 20 per cent” at hats, but at shoes “he can excel him by one-third or 33 per cent.” Both gain if the better man makes only shoes and the other only hats.
Three implications follow:
- You do not need to be the best in an arena to belong there. You need your edge there to be larger than it is in your other options.
- Being good at many things is not a reason to compete in all of them. The better shoemaker still loses value by making hats.
- The comparison is with your own alternatives. The question is not “am I good at this?” but “is this where my edge is largest?”
Modern labour economics uses the same logic. Daron Acemoglu and David Autor’s 2010 NBER chapter proposes a task-based model in which “the assignment of skills to tasks is endogenous” and technical change can mean “the substitution of machines for certain tasks previously performed by labor.” Which tasks a person ends up doing depends on relative advantage, and technology shifts that assignment. AI is the current shift.
David Deming’s NBER working paper on social skills (revised June 2017) models workers who “trade tasks” to exploit comparative advantage, with social skills lowering the cost of that trade. His employment data are the closest thing to evidence that combinations, not single skills, define the best arenas.
| Finding | Figure as reported | Period and data | Source |
|---|---|---|---|
| Jobs high in both math and social skill intensity | grew by about 7.2 percentage points as a share of the US labour force | 1980-2012, US | Deming, NBER WP 21473 |
| Jobs low in math, high in social skill intensity | grew by about 4.6 percentage points | 1980-2012, US | Deming |
| All social skill-intensive occupations | total increase of 11.8 percentage points in employment share | 1980-2012, US | Deming |
| Jobs high in math but low in social skill intensity | shrank by about 3.3 percentage points | 1980-2012, US | Deming |
| University graduates’ wage growth attributed to task-specific human capital | at least 40 percent over a ten-year period | Germany | Gathmann and Schönberg, IZA DP 3067 |
| Same measure, low- and medium-skilled workers | at least 35 and 25 percent | Germany | Gathmann and Schönberg |
Sources: Deming, NBER Working Paper 21473 (revised June 2017), text accompanying Figure 4; Gathmann and Schönberg, IZA Discussion Paper 3067 (2007), abstract.
The Gathmann and Schönberg result adds a constraint: skills travel, but mostly to nearby arenas. Their abstract reports that “individuals move to occupations with similar task requirements.” An arena you can reach with your existing task capital is cheaper to enter than one that resets you to zero. The published piece on skill stacking covers how combinations create rarity; arena selection decides which market that rarity is sold into.
Why do some arenas pay the top player far more than the runner-up?
Arenas differ in payoff shape, not only in size. The classic explanation is Sherwin Rosen’s 1981 paper “The Economics of Superstars”. As summarised by Douglas Clement for the Federal Reserve Bank of Minneapolis (2002), Rosen explained very large top incomes with two forces: “imperfect substitution among different sellers and scale economies of joint consumption technologies.” Lesser talent is a poor substitute for greater talent (“Hearing a succession of mediocre singers does not add up to a single outstanding performance,” in Rosen’s words as quoted there), and some technologies let a few sellers serve the whole market.
The same article quotes a later theory by the economists Boldrin and Levine predicting that “the increased reproducibility of information will continue generating large income disparities among individuals of very similar skills and in a growing number of industries.” It is a hypothesis, quoted in 2002 about digital copying, not a measured result. But it describes the direction AI pushes: when output is cheap to reproduce, the reward for being the chosen seller grows relative to the reward for being a competent one.
Payoff shape changes the question you ask before entering. In a narrow-spread arena, being solidly good pays roughly like being very good. In a wide-spread arena, the realistic question is whether you can reach the top fifth within a few years, because the middle pays less than the averages suggest.
What does the arena data actually show?
CEOtudent joined two BLS datasets that are published separately:
- BLS Employment Projections, AI exposure categories and 2025-35 projections: for 831 detailed occupations, 2025 employment, projected change to 2035, annual openings, percent self-employed, median wage and a four-level “relative AI exposure” category built from five external sources.
- BLS Occupational Employment and Wage Statistics (OEWS), May 2025: wage percentiles for the same occupation codes. The ratio of the 90th to the 10th percentile (P90/P10) is used here as a simple measure of how widely pay spreads inside an arena.
830 of the 831 occupations matched on code. BLS is explicit about the limits: “An exposure category is not a forecast of employment growth or decline,” its theoretical sources “conceptualize AI capabilities as those available no later than mid-2023,” and the categories “do not distinguish between AI impacts from automation versus augmentation.”
Table 1. Arena conditions by BLS relative AI exposure category (CEOtudent analysis of BLS Employment Projections AI exposure categories, 2025-35, and BLS OEWS May 2025)
| BLS relative AI exposure | Occupations | Employment 2025 | Projected change 2025-35 (employment-weighted) | Openings per 100 workers per year | Median P90/P10 inside occupation | Median of occupation median wages |
|---|---|---|---|---|---|---|
| Low | 213 | 30.3 million | 2.5% | 10.8 | 1.93 | $49,120 |
| Moderate | 206 | 44.7 million | 6.2% | 13.2 | 2.17 | $52,760 |
| High | 206 | 42.1 million | 3.1% | 9.0 | 2.47 | $73,985 |
| Very high | 206 | 53.2 million | 2.0% | 8.4 | 2.70 | $78,105 |
Employment, projected change and openings are from the BLS AI exposure table; P90/P10 ratios are computed from OEWS May 2025 national percentiles (hourly percentiles for the few occupations published hourly only; 212 matched ratios in the Low row). Derived cells are recomputed in a script.
Three things stand out. Pay spread rises with every step of exposure, from 1.93 to 2.70: the most AI-exposed arenas are the ones where top earners pull furthest from the bottom of the same occupation. Exposed arenas pay more but offer less room: Very high exposure occupations have the highest typical wages and the lowest growth and openings rate, while the Moderate group, much of it hands-on work, has the most room on both. Openings are mostly replacement, not growth: of 17,461.8 thousand openings a year across all occupations, 16,870 thousand (96.6 percent) come from people leaving the occupation or the labour force. A high openings rate often signals churn.
The correlation needs care. High and Very high exposure occupations are also better-paid, more credentialed occupations, where long seniority ladders naturally widen pay. Table 1 does not show that AI causes wider spreads. It shows that the arenas where AI is most applicable today are also the most tournament-like, exactly the setting where position matters more than effort.
The exposure label also says little about any one arena. The next table holds it constant.
Table 2. Same exposure label, very different arenas (CEOtudent analysis of BLS Employment Projections 2025-35 and BLS OEWS May 2025)
| Occupation (arena) | BLS AI exposure | Employment 2025 (thousands) | Projected change 2025-35 | Openings per 100 workers per year | P90/P10 | Self-employed, 2025 |
|---|---|---|---|---|---|---|
| Data scientists | Very high | 275.6 | 34.6% | 9.0 | 2.96 | 1.9% |
| Information security analysts | Very high | 192.9 | 21.0% | 7.3 | 2.66 | 0.2% |
| Management analysts | Very high | 1,077.1 | 10.1% | 8.7 | 2.83 | 14.0% |
| Software developers | Very high | 1,717.8 | 10.2% | 5.5 | 2.60 | 1.5% |
| Project management specialists | Very high | 1,094.3 | 6.7% | 7.0 | 2.73 | 1.2% |
| Market research analysts and marketing specialists | Very high | 952.7 | 7.0% | 8.6 | 3.58 | 3.8% |
| Accountants and auditors | Very high | 1,595.2 | 5.0% | 7.2 | 2.57 | 5.5% |
| Lawyers | Very high | 863.7 | 4.7% | 3.3 | 4.49 | 11.3% |
| Personal financial advisors | Very high | 299.4 | 1.4% | 5.7 | 7.11 | 9.7% |
| Public relations specialists | Very high | 314.5 | 1.8% | 7.3 | 3.06 | 1.1% |
| Writers and authors | Very high | 140.3 | -0.3% | 8.5 | 3.16 | 64.8% |
| Editors | Very high | 105.0 | -1.1% | 7.5 | 3.73 | 9.0% |
| Graphic designers | Very high | 253.1 | -1.7% | 6.3 | 2.65 | 20.5% |
| Nurse practitioners (contrast) | High | 336.3 | 41.0% | 8.7 | 1.72 | 1.6% |
| Electricians (contrast) | Moderate | 821.0 | 9.2% | 8.9 | 2.54 | 7.1% |
Selection: common knowledge-work arenas plus two contrasts, chosen by the editors for relevance, not a ranking. OEWS covers wage and salary workers only, so P90/P10 understates spread where many work freelance (writers and authors: 64.8 percent self-employed).
Thirteen arenas carry the same “Very high” label, yet projected growth runs from -1.7 percent (graphic designers) to 34.6 percent (data scientists), and pay spread from 2.57 (accountants and auditors) to 7.11 (personal financial advisors). Two people who both “work in an AI-exposed field” can be in arenas with opposite dynamics. The label says the work will change. Room and payoff shape say whether the arena is worth entering.
What kinds of arena are there?
Splitting all 830 matched occupations on projected growth (above or below the all-occupation 3.5 percent) and P90/P10 (above or below the median of 2.29) gives four arena types.
Table 3. Four arena types (CEOtudent editorial framework applied to a CEOtudent analysis of BLS Employment Projections 2025-35, AI exposure categories, and OEWS May 2025)
| Arena type | Rule | Occupations | Employment 2025 | Share of that employment in High or Very high AI exposure | Three largest by employment | What it rewards |
|---|---|---|---|---|---|---|
| Open field | Growing, narrow spread | 143 | 42.6 million | 28% | Home health and personal care aides; fast food and counter workers; registered nurses | Reliable competence and credentials; upside is capped |
| Contested frontier | Growing, wide spread | 213 | 43.1 million | 72% | General and operations managers; software developers; accountants and auditors | A clear position; room exists, but the top pulls away |
| Commodity arena | Slow, narrow spread | 272 | 58.1 million | 54% | Retail salespersons; cashiers; laborers and freight, stock, and material movers, hand | Cost and reliability; little reward for being exceptional |
| Tournament arena | Slow, wide spread | 202 | 26.4 million | 80% | Waiters and waitresses; first-line supervisors of office and administrative support workers; first-line supervisors of retail sales workers | Rank; only a top position pays well |
Thresholds: 3.5 percent is the BLS projected change for total employment, 2025-35 (Employment Projections Table 1.2); 2.29 is the median P90/P10 across the 830 matched occupations. Type names and the “what it rewards” column are a CEOtudent editorial framework, not BLS categories.
Two readings matter. The Tournament arena is where AI exposure is most concentrated: 80 percent of its employment is in High or Very high exposure occupations. In a slow-growing, wide-spread arena, the realistic plan is to climb or to reposition, not to hold the middle. And the Contested frontier, where most growth-seeking knowledge workers will play, is a positioning game: room exists, but the reward goes to a distinct position, not general competence.
The Arena Selection Scorecard
Score each candidate arena 0, 1 or 2 on six tests. Define the arena narrowly enough that you could name ten competitors.
Table 4. The Arena Selection Scorecard (CEOtudent editorial framework)
| # | Test | 0 | 1 | 2 | Logic |
|---|---|---|---|---|---|
| 1 | Relative edge | Edge no larger than in your other options | Somewhat larger | Clearly where your edge over the typical entrant is largest | Ricardo |
| 2 | Payoff-shape fit | Wide spread and no credible path to the top fifth in 3-5 years | Unclear | Narrow spread, or wide spread with a credible path up | Rosen |
| 3 | Room | Projected decline and few openings | Flat, or growth carried by churn | Growth above the economy-wide figure and real openings | BLS projections |
| 4 | AI direction | AI already produces the core output you would sell | Mixed | AI handles the routine parts; your edge is what remains (judgment, relationships, accountable decisions, combination) | BLS exposure; Acemoglu and Autor |
| 5 | Trade-off clarity | Cannot name what you would stop doing | Named, not committed | A written list of clients, tasks and channels you will decline | Porter |
| 6 | Adjacency | No overlap with current tasks | Partial overlap | Most current task capital carries over, or a short bridge exists | Gathmann and Schönberg |
Reading the total out of 12 (CEOtudent editorial thresholds):
- 10-12: commit. Make it the primary arena and build the learning plan around its specific edge.
- 7-9: enter with a test. Run a bounded experiment first: a project, a pilot client, a portfolio piece. The piece on the explore-exploit dilemma covers how to size and stop such experiments.
- 6 or below: do not enter as designed. Narrow the arena until the edge grows, or pick another.
A zero on test 1 or test 5 blocks a commit decision whatever the total. Without a relative edge there is no position; without a trade-off there is no strategy, only a wish list.
How to run the scorecard in one sitting
- List three candidate arenas, narrowly. Not “data” but “pricing analytics for subscription software companies”; not “writing” but “technical documentation for regulated medical devices”.
- Pull the public numbers for the nearest occupation. BLS projections, openings, exposure and OEWS percentiles exist for every detailed US occupation; elsewhere, use the national statistics office and note what is missing. Table 2 shows the columns worth recording.
- Score test 1 honestly. Ask two people who have seen your work where you outperform peers, and write the comparison with your own alternatives down.
- Check payoff shape against your likely rank. If P90/P10 is above roughly 3 (a CEOtudent rule of thumb, not a BLS threshold) and you are not yet in the upper part of that arena, test 2 scores 0 or 1 until there is evidence of a path up.
- Write the not-to-compete list. Clients you will not take, tasks you will not sell, platforms you will not chase. The piece on what not to learn applies the same discipline to learning time.
- Turn the winning arena into one learning target. This is the student half: the edge that separates the top fifth from the median in that arena. To choose between candidate skills once the arena is fixed, use the 4-factor skill decision matrix.
- Set a review date. BLS publishes projections annually; the 2026-36 projections are scheduled for release in 2027.
What should you refuse to compete on?
Porter’s argument is that trade-offs make a position sustainable: “a trade-off means that more of one thing necessitates less of another.” In AI-exposed arenas, three refusals tend to matter:
- Generic throughput. If every competitor has the same AI assistants, speed on standard deliverables converges towards the market price.
- The median of a tournament arena. Where spread is wide and growth slow (Table 3), the middle is the expensive option, not the safe one.
- Arenas defined by a tool. A tool is not a market. An arena is defined by who pays and for what problem; the tools inside it will change many times.
Choosing what not to do also frees the student: a narrow arena tells you which knowledge can decay harmlessly and which must compound. The piece arguing that the T-shaped professional is obsolete describes how the shape of that knowledge is changing; arena choice tells you where to grow it.
What this does not mean
- Not avoiding AI-exposed arenas. BLS states that “High” or “Very high” exposure “does not necessarily mean employment will decline.” Several Very high arenas in Table 2 are projected to grow faster than the economy.
- Not chasing the fastest-growing occupation. Growth without a relative edge is a crowded entrance.
- Not reading P90/P10 as talent. It is a spread of wage-and-salary pay that also reflects seniority, region, employer size and hours, and it excludes the self-employed.
- Not fixed categories. The arena-type thresholds come from one projection cycle; occupations near a line can move when data are revised.
- Not one arena for life. Strategy is a choice made with current information and reviewed on a schedule.
This piece is general information, not personalised career, financial or legal advice.
Frequently asked questions
What is strategic positioning for an individual?
The deliberate choice of where to compete (a specific occupation, client type or problem) and how to win there (the edge that makes someone pick you over the next qualified person), including the choice of what not to do.
How is this different from building career capital?
Career capital is the assets you accumulate. Arena selection is the market where those assets are sold. The same skills can be worth much more in one arena than another, so the arena choice comes first.
Does a “Very high” AI exposure rating make an arena a bad choice?
No. BLS describes the categories as relative and not employment forecasts. In the CEOtudent analysis, Very high exposure arenas have the highest typical pay but the widest spread and slowest average growth, so they reward a clear position more than average competence.
What does P90/P10 say about an arena?
It is the 90th-percentile wage divided by the 10th-percentile wage inside one occupation. Higher means rank matters more. Across 830 occupations the median is 2.29; personal financial advisors are at 7.11 and nurse practitioners at 1.72.
What should someone who is good at several things do?
Apply Ricardo’s logic: compare the edge across your own options and specialise where it is largest relative to the typical entrant. Then check that the arena has room and a payoff shape that fits the rank you can reach.
How often should an arena choice be re-scored?
Once a year is a reasonable default, timed to new data such as the annual BLS projections, and immediately after a large change in your edge, your field or the tools that do your core work.
Sources
- U.S. Bureau of Labor Statistics, Employment Projections. AI exposure categories and 2025-35 employment projections (supplemental table, with Field Descriptions sheet).
- U.S. Bureau of Labor Statistics. Artificial Intelligence (AI) exposure categories: methodology and limitations.
- U.S. Bureau of Labor Statistics, Employment Projections 2025-35. Occupation tables 1.2 (projections and worker characteristics) and 1.10 (separations and openings).
- U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics. May 2025 national cross-industry estimates and field descriptions.
- Porter, Michael E. What Is Strategy? Harvard Business Review, November-December 1996.
- Martin, Roger L. The Big Lie of Strategic Planning. Harvard Business Review, January-February 2014.
- Ricardo, David. On the Principles of Political Economy and Taxation, third edition, 1821, chapter 7, On Foreign Trade. Library of Economics and Liberty edition.
- Clement, Douglas. Perfectly Competitive Superstars. Federal Reserve Bank of Minneapolis, 2002. Source for its summary of and quotations from Rosen (1981), The Economics of Superstars, and from Boldrin and Levine; the original papers were not opened.
- Deming, David J. The Growing Importance of Social Skills in the Labor Market. NBER Working Paper 21473, 2015, revised June 2017.
- Gathmann, Christina, and Uta Schönberg. How General Is Human Capital? A Task-Based Approach. IZA Discussion Paper 3067, 2007 (abstract and introduction).
- Acemoglu, Daron, and David Autor. Skills, Tasks and Technologies: Implications for Employment and Earnings. NBER Working Paper 16082, 2010 (abstract).
Editorial note: Tables 1, 2 and 3 are CEOtudent analyses joining the BLS Employment Projections AI exposure table (2025-35) with BLS OEWS May 2025 wage percentiles by occupation code; every derived cell is recomputed in a script. The arena-type names and the Table 4 scorecard with its thresholds are a CEOtudent editorial framework. Popular claims that a fixed percentage of jobs will be replaced by AI were deliberately not used: BLS states its exposure categories are not employment forecasts, and no primary source opened for this piece supports such a figure. Rosen (1981) and Boldrin and Levine are cited only as summarised by the Federal Reserve Bank of Minneapolis.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:















