The testimony published by Wired offers a rarely concrete glimpse behind the scenes of regenerative artificial intelligence. The author, a Hollywood screenwriter and showrunner, recounts how, after the 2023 strike and the sustained slowdown of the American visual industry, she turned to AI model training contracts to pay her rent and living expenses. What was presented to her as easy money quickly turned into a series of unstable, poorly supervised, and often exhausting tasks.
Her account sheds light on a reality often forgotten in enthusiastic discussions about AI: behind conversational bots, image generators, and recommendation systems, thousands of human workers evaluate, annotate, correct, classify, and test the responses produced by the models. They must judge the tone of an assistant, write precise scripts for videos, identify objects in images, draft test scenarios, or spot security flaws. All in an environment where speed, constant availability, and obedience to instructions seem to matter as much as expertise.
The screenwriter mentions working for several platforms specialized in this market, including Mercor, Outlier, Turing, Handshake, and Micro1. The initial promises may have seemed attractive, with high hourly rates for qualified profiles. But, according to her experience, contracts follow one after another without guarantee, projects start late, end without notice, and workers can be excluded from a Slack channel or platform from one day to the next. In this system, they are not considered employees but as “taskers,” executing micro-tasks.
One of the most revealing passages concerns the contradiction at the core of this model. Platforms boast the freedom to work when and as much as one wants, but in practice, tasks can appear at any time, be limited in number, and disappear quickly. Workers must stay connected, monitor their emails, Slack, and internal dashboards, sometimes late at night or in the middle of the night. Those who don’t react quickly enough risk earning nothing.
The testimony also describes a highly hierarchical system but without real stability. Very young team leaders supervise professionals often older and more experienced, from cinema, television, teaching, journalism, or other vulnerable sectors. Instructions change, evaluation criteria remain vague, grades drop without a clear explanation, and workers must constantly undergo new, often unpaid training. The motivation vocabulary, badges, scores, rankings, enthusiastic messages, and calls to “finish strong” barely conceal the precariousness of the situation.
This story addresses a broader issue: AI not only replaces some workers, but it also transforms others into invisible labor for machines. In Hollywood’s case, the irony is harsh. Screenwriters who went on strike to prevent studios from replacing them with AI find themselves training systems that could further undermine their profession a few months later. Their narrative expertise, language skills, and scene judgment become resources used to enhance automated models.
It is also mentioned that this economy is based on a large asymmetry. AI companies and their subcontractors can quickly mobilize thousands of independent workers, adjust rates, interrupt projects, and shift tasks from one platform to another. Workers, on the other hand, bear uncertainty, unpaid waiting periods, tool changes, availability requirements, and lack of protections usually associated with employment. Wired reports that several lawsuits have been filed against Mercor for allegedly misclassifying some workers as independent contractors.
Beyond the personal story told in Wired, this testimony raises a central question for the future of digital work: how much automation can be done without recognizing the human work that makes this automation possible? AI models are often presented as autonomous systems capable of learning from massive data volumes. However, their refinement still largely depends on people who evaluate, correct, and guide their behavior, sometimes under conditions close to a constant race.
For cultural industries, the signal is particularly worrisome. AI doesn’t just arrive as a tool for creation or productivity but inserts itself into a precarious labor market where qualified professionals accept unpredictable contracts because their original sector no longer offers enough security. The risk is not only that machines will produce more content but also that creative professions will break down into fragmented, timed, evaluated, and underpaid tasks.
The Wired testimony has the power of a personal story but goes far beyond the case of an American screenwriter. It reveals an AI economy built on a promise of modernity while echoing very old reflexes of precarious work: constant availability, lack of security, performance pressure, and unilateral power of the employer or client. Behind the smooth image of digital assistants, there are still humans. And too often, it is they who the system makes replaceable.
Source: Wired
