09 Work & Economy 11 min read
Complement or Substitute
Comparative advantage guarantees there will be something for you to do. It guarantees nothing whatsoever about what it pays.
Two arguments dominate this subject and both are lazy. One says technology has always created more jobs than it destroyed, so relax. The other says this time the machines take everything, so despair. The first is a historical generalisation being asked to do predictive work it cannot do. The second assumes a conclusion. The useful frame is narrower, older, and comes from labour economics: automation does not act on jobs. It acts on tasks.
Tasks, not jobs
Every job is a bundle of tasks. A technology arrives and does some of them. What happens to the wage depends entirely on the relationship between the automated tasks and the remaining ones.
If the machine substitutes for your tasks, your labour is worth less. If it complements them — makes the tasks you still do more productive or more valuable — your labour is worth more. The same technology does both, to different people, simultaneously. This is why “will AI take jobs?” has no answer and “which of my tasks does it substitute and which does it complement?” has a very concrete one.
The canonical illustration is the ATM. Introduced from the 1970s, it automated the core task of a bank teller — dispensing cash. Teller employment in the United States then rose for roughly three decades. Branches became cheaper to operate, banks opened more of them, and the surviving teller role shifted toward sales and relationship work that the machine complemented rather than replaced.
What is genuinely new this time
Previous automation waves hit routine work — first routine manual (assembly), then routine cognitive (clerical, bookkeeping). This produced the well-documented polarisation of labour markets: growth at the high-skill end, growth at the low-skill service end, hollowing in the middle.
The standard advice that followed was: move up. Get educated. Do non-routine cognitive work — analysis, writing, design, diagnosis, strategy. An entire generation was directed into that harbour by every careers service in the developed world.
That harbour is what this wave hits first. It is the substantive novelty and it should not be minimised: the escape route from the last disruption is the frontline of this one.
The horse analogy is imperfect in a way that matters and is usually stated badly. Humans differ from horses in two respects that are decisive: we are the consumers whose demand constitutes the economy, and we are the voters who set the rules of distribution. Neither guarantees a good outcome. Both mean the outcome is political rather than mechanical — which is a more useful thing to know than either of the lazy arguments.
Where the value goes
When one input to production becomes abundant and cheap, value accrues to whatever remains scarce and is required alongside it. If intelligence is becoming the cheap input, the question is: what are its scarce complements?
| Scarce complement | Why cheap intelligence raises its value |
|---|---|
| Judgment and taste | Infinite competent output makes selection the bottleneck. Someone must decide which of a thousand plausible options is right. |
| Trust and provenance | When anything can be fabricated, a verifiable track record becomes infrastructure rather than a nicety. |
| Physical presence and dexterity | Robotics lags language by a wide margin. Plumbing, care, surgery, repair, construction — the world is stubbornly physical. |
| Accountability | Someone has to be answerable. A model cannot sign. See responsibility. |
| Distribution and relationships | Access to customers, patients, clients, an audience. Cheap production makes attention the constraint. |
| Licences and legal standing | Regulated gates move slowly and deliberately. This is a real moat and an honest one to name. |
| Energy, compute, data, capital | The physical substrate of the abundance. Whoever owns it captures a large share of what it produces. |
That last row is the uncomfortable one. If output is increasingly produced by capital rather than labour, income flows to capital owners, and no amount of individual skill development changes that. It is a distributional question, not a career question, and it belongs on the civilization page — but it should be named here, because career advice that ignores it is selling personal solutions to a structural problem.
Positioning without forecasting
Nobody knows the timeline. Anyone who tells you they do is selling something. What you can do is hold a position that pays off under several futures at once.
- Domain depth plus tool fluency. The most reliably valuable combination available. Not a technologist, not a domain expert — the domain expert who understands the tools better than anyone else in the building. This is the education barbell in career form.
- Get closer to consequence. Move toward the end of the chain where outcomes are felt and someone has to answer for them. Work that terminates in a deliverable is easier to automate than work that terminates in a decision somebody lives with.
- Own something. A stake, an audience, a client relationship, equity, a practice. Pure wage labour is the most exposed position in a capital-intensive transition, and it is the default position.
- Build verifiable trust. Track record, references, professional standing. Slow to build, impossible to fabricate, appreciating.
- Take the physical seriously. The reflex to regard embodied work as a lesser option is a status habit from the last era, and it is about to look expensive.
Correct, and it applies to the whole genre. Individual positioning is real and it is also a rearrangement of who absorbs the shock, not a reduction in the shock. Both things need doing: individuals should position, and the distributional question has to be answered politically, because it will not answer itself. Pretending the first substitutes for the second is the characteristic dishonesty of writing about this subject — including, without constant vigilance, this site.
Where work sits on the ladder
- 1
Cede Machine
The machine does it. You never see the work.
Tasks you cede entirely are tasks you are no longer paid for. Map them honestly — this is your exposure.
- 2
Approve Machine-led
The machine drafts. You review and sign.
- 3
Collaborate Shared
You and the machine work the problem together.
Where the productivity gains are largest. Also where the largest measured gains go to the previously lowest performers, which compresses the wage premium for competence.
- 4
Critique Human-led
You do the work first, unaided. Then ask the machine to attack it.
- 5
Reserve Human
You do it alone, on purpose, knowing help exists.
Economically irrational in the short run, and the only thing that keeps you employable when rung three stops needing you.
List what you actually did last week in units of thirty minutes. Mark each task substitute, complement or neutral. Your exposure is the substitute column as a share of your time — a far more honest number than any occupation-level forecast.
Take on more of what the tools make more valuable: the judgment calls, the client relationship, the accountable decision, the physical execution. Do this before you are forced to.
Something embodied, licensed or relational. Not as a fallback — as portfolio diversification against a correlated shock to cognitive labour.
Equity, clients, an audience, a practice, ownership of any kind. Even a small stake changes which side of the capital/labour split you sit on.
How the gains are distributed will affect your life more than your skill choices will. Career strategy that treats distribution as somebody else’s department is incomplete strategy.