Prompt Engineering Is Piece-Rate

In the English textile industry of the late 17th century, there was a system called the putting-out system. A merchant capitalist delivered raw wool to a cottager. The cottager, working in their own home, with their own loom, on their own schedule, spun the wool into yarn. The merchant collected the yarn and paid a piece rate. The cottager was told, in every period pamphlet, that this was an entrepreneurial opportunity, a way to be one’s own boss, a liberation from the grinding servitude of the old guilds.

The cottager was paid, in real terms, less than the guild spinner had been paid. The cottager absorbed the cost of the loom, the cottage, the heat, the light, and the food. The merchant absorbed no overhead and took all the surplus. When demand fell, the merchant simply stopped delivering wool. The cottager starved.

The modern version of the putting-out system is called Fiverr, Upwork, the Contra marketplace, the Anthropic “solutions partner” program, and the OpenAI “custom GPT builder” store. The wool is a prompt. The yarn is a PDF or a marketing plan or a logo. The cottage is your apartment. The loom is a ChatGPT Plus subscription you pay for yourself. The piece rate is whatever the algorithm says it is this week.

Prompt engineering is not a trade. It is a putting-out system with a LinkedIn banner.

The anatomy of a piece-rate job

A piece-rate job, in political-economy terms, has five features. It is worth listing them plainly, because they are obscured in every contemporary pitch of the “AI freelancer lifestyle.”

1. The worker owns the means of production. The loom, the cottage, the ChatGPT subscription, the laptop, the insurance, the electricity, the tax prep software.
2. The worker bears the risk. A slow week, a lost client, an algorithm change, a model-capability jump that makes the task obsolete — all of these are the worker’s problem.
3. The rate is set by the buyer, not negotiated. There is a market rate. You take it or you take no work. The individual seller has no pricing power against a platform aggregating millions of buyers.
4. There is no continuity of employment. Each piece is a standalone contract. Benefits, notice periods, severance, pension contributions — all zero.
5. The worker is told they are an entrepreneur. The pamphlet exists in every century. It is always false in the same way. A real entrepreneur owns the customer relationship and the product specification. A putting-out worker owns neither.

A 2026 Upwork AI specialist has all five features. A 1680 Cotswolds cottager had all five features. The technology has changed. The labour relation has not.

”Be your own boss” is the oldest lie in labour history

The single most persistent slogan of piece-rate capitalism across four centuries is “be your own boss.” It has been deployed by the English wool merchant, the French silk merchant, the Amazon Mechanical Turk program, Uber, DoorDash, TaskRabbit, and now every AI-freelancing platform in the current generation.

The slogan survives because it contains a fragment of truth. The piece-rate worker does have more scheduling autonomy than a wage worker. The piece-rate worker is not watched by a supervisor. The piece-rate worker can, in theory, turn down a piece of work. These are real differences.

The differences conceal a more important similarity. The piece-rate worker and the wage worker are both in a position where the cost structure of the buyer is not their concern and the cost structure of the worker is not the buyer’s concern. The worker absorbs every cost shift. The buyer absorbs no cost shift. Over a long enough horizon, the distribution of the surplus trends hard in the buyer’s favour. This is true whether the buyer is a 17th-century merchant or a 21st-century marketplace. Autonomy, such as it is, is the consolation prize.

The AI-specific twist

What makes the AI version of the putting-out system novel, and worse than the 17th-century original, is the pace at which the buyer’s alternative to the worker improves.

In the 17th century, the alternative to the cottage spinner was a different cottage spinner, or the factory (which did not yet exist at industrial scale). The buyer’s alternative improved slowly, over generations. The piece rate eroded slowly. The worker had a generation to adapt.

In 2026, the alternative to the AI freelancer is a model that is six months away from doing the same task at one-tenth the cost. Every time the model improves, the entire class of freelancers specializing in the newly-automated task becomes instantly redundant. The copywriter who specialized in SEO meta descriptions in 2022 was redundant by 2024. The junior illustrator who did rapid concept art was redundant by 2023. The voice-over talent who did corporate training was redundant by 2024. The transcription specialist was redundant by 2022.

Each wave of redundancy is absorbed individually by the affected workers. Each worker is told to retrain, move to a new specialization, stay ahead of the curve. The new specialization is redundant a year later. The worker is on a treadmill that accelerates every quarter.

This is not an employment relationship. This is a race against a deflating commodity. The commodity is the worker’s specific skill. The deflator is the model. The worker cannot win, because the model’s improvement rate is higher than the worker’s learning rate, and capital has endless budget to fund the model’s improvement.

The Canadian freelancer and the CRA

The Canadian twist on this story is the Canada Revenue Agency. A Canadian freelancer is a self-employed person for tax purposes. The CRA demands quarterly instalments, collects GST on invoices above $30,000 a year, and offers a thin deduction regime that does not meaningfully offset the cost of being one’s own employer.

The typical Canadian AI freelancer making $50,000 a year gross is paying the full CPP contribution on both halves (employer and employee), self-funding EI they cannot claim in most circumstances, remitting GST they then fight to collect from clients, and filing a T2125 that the CRA routinely audits. The net income, after overhead, is roughly comparable to a $38,000 salaried job with benefits.

The Canadian state has, in effect, designed its social insurance system around the 20th-century industrial employment model, and has failed to adapt that model to the new piece-rate majority. The consequence is that the fastest-growing class of Canadian workers — freelancers, contractors, gig workers, AI-adjacent specialists — are paying a higher effective rate into the social insurance system while being structurally excluded from most of its benefits.

This is a regressive transfer. It has not been acknowledged by any federal government in the current parliamentary cycle.

What a piece-rate labour movement looks like

It is possible to organize piece-rate workers. It has been done. The historical case studies are instructive.

The Lyon silk weavers organized twice in the early 19th century, in 1831 and 1834. They won a minimum piece rate through city-level negotiation, lost it to a military crackdown, and established a tradition of mutual-aid societies that outlived every Napoleonic regime. The organizing vehicle was geographic, not workplace-based. The demand was a rate floor, not a wage.

The American hand-shoemakers, in the 1830s and 1840s, organized the Knights of St. Crispin, which at its peak had 50,000 members, all of them piece-rate cottage workers. The union won rate floors through coordinated refusal to work below a set price. The union collapsed when industrial mechanization made the shoes cheaper to produce than the piece rate could be sustained.

The modern analogue is not yet clear. The most promising experiments are platform-worker unions — the Independent Workers’ Union of Great Britain, the California Rideshare Drivers United group, the emerging AI Freelancers Collective in Germany. None of these has cracked the scale problem. The platform worker is geographically dispersed, the platform itself is the employer in economic substance but not in legal form, and the regulator is two steps behind in every jurisdiction.

A Canadian piece-rate labour movement for AI freelancers would need three things that do not currently exist. A legal status between “employee” and “independent contractor,” which Quebec has attempted but the federal government has not adopted. A platform-worker bargaining framework, which the current Labour Code does not provide for. And a sectoral union body willing to organize across platforms, which no Canadian union has yet committed to.

The argument

Prompt engineering is not a craft. Prompt engineering is the current name for the last human touch in a production pipeline that is being progressively automated. The name will change in two years. The role will be absorbed in five. The workers currently doing it are piece-rate labour, wearing the same pamphlet the 17th-century cottager wore, and carrying the same risks.

The exit from piece-rate is organization. The organization of piece-rate workers is hard but has been done before. The historical precedent is not American mid-century unionism. It is 1830s Lyon and the Knights of St. Crispin. Read them. Borrow the tactics. Pay the dues.

The putting-out system lost in the 19th century because the workers organized against the rate cuts that the system depended on. The AI putting-out system of the 2020s will lose or continue for the same reason. It is a fight about the rate. It has always been a fight about the rate.