In February 2026, April Watson hit her head in an Amazon warehouse in the United States. What followed was not the intervention of a supervisor nor the activation of a medical protocol with a human face. An artificial intelligence assistant could not locate the correct workplace accommodation form while the Time Off Task counter kept running. The counter does not know there is an injury. It cannot know. It was designed to measure scanner inactivity, not the state of the body holding it. This distinction, absent from the system, is not a technical oversight. It is the premise on which the entire architecture of algorithmic work management in 2025-2026 operates. The worker’s body exists as a source of data, not as a subject of rights. What the system records is not a person. It is a terminal.
Time Stolen, Minute by Minute
The mechanism is straightforward, as all well-designed extraction mechanisms are. Amazon calls Time Off Task, TOT, the period during which a worker’s scanner remains inactive. Internal documents submitted to the National Labour Relations Board in the United States reveal the detail. A written warning for accumulating thirty minutes of inactivity in a single day; automatic dismissal upon reaching one hundred and twenty minutes in a single shift, or thirty minutes across three separate days within a year. The system maintains a record minute by minute. Toilet breaks are included in the count. A worker dismissed in 2019 received from his supervisor a sheet listing every exact interval of inactivity during his shift, so he could explain where he had been. The explanation changed nothing. The algorithm had already decided.
What makes this mechanism something more than an anecdote of corporate cruelty is its internal logic. The TOT does not distinguish between the worker who hides his phone behind the boxes and the worker who cannot walk because his knee is in pain. It does not distinguish between resistance and incapacity. It was not built to make that distinction because the distinction does not interest the system. The body appears in that equation as a dependent variable. If it produces at the expected rate, it is invisible. If it does not produce, it becomes a negative data point that triggers an automated sequence of warnings, then sanctions, then expulsion. Nobody signed the dismissal order. The system produced the result.
According to the report of the US Senate Committee on Health, Education, Labour and Pensions, published in December 2024, Amazon warehouse workers suffer rates of serious injury that nearly double the average for the American logistics industry. The report documents a direct correlation between productivity tracking systems and the increase in musculoskeletal injuries.
The Fiction of the Autonomous Collaborator
In the other modality of platform work, the violence has a different vocabulary but an identical structure. The Deliveroo rider, the Uber Eats driver, the Rappi courier is not an employee. He is a collaborator, an independent entrepreneur who uses his own vehicle, assumes his own risks, and organises his time freely. This legal fiction has precise material consequences. No social security, no accident compensation, no human interlocutor when the algorithm blocks the account. The collaborator’s autonomy is real in one sense only. The company transfers to him the totality of the costs and risks of the operation whilst retaining complete control over the conditions under which it is carried out.
The paradigmatic case is Deliveroo’s Frank algorithm, condemned by the Bologna tribunal in December 2020. Frank assigned work slots according to a performance ranking system. Riders who missed reserved slots saw their score fall and received fewer orders. Frank did not record the reason for the absence. An absence due to strike produced the same effect as an absence without notice. The tribunal fined Deliveroo fifty thousand euros and declared that the system’s blindness to the reasons for absence constituted discrimination. The company modified the algorithm days before the ruling. The principle of selective blindness remains, however, as a design norm across the entire industry. In November 2025, Worker Info Exchange filed a class action at the Amsterdam District Court against Uber’s dynamic pay-setting system, arguing that the algorithmic model has systematically reduced driver earnings since 2020 through opaque decisions over which workers have neither information nor recourse. In Colombia, some 645,000 active Rappi delivery accounts in 2024 operated with a base rate of 3,400 pesos per delivery. Law 2466, enacted in June 2025, for the first time mandates algorithmic transparency and the right to human review of any automated decision.
Blindness as Architecture
Blindness is an architectural decision, not an implementation defect. A human foreman in a textile factory in the nineteen-fifties could distinguish between the worker feigning clumsiness and the worker carrying an undeclared fracture. That capacity for distinction was inconvenient for capital because it created zones of negotiation, spaces of humanity within the extraction relationship. The human manager could be corrupted, persuaded, moved. The algorithm cannot be, and that impossibility is not a technical limitation. It is its fundamental operational advantage. A system that does not distinguish between resistance and incapacity cannot be accused of malice. It eliminates malice along with mercy. What remains is a process.
The most advanced form of this logic no longer measures the gesture but the body itself. In industrial warehouses across Asia, Europe and North America, portable biometric monitoring devices record in real time the heart rate, skin temperature, galvanic response, and movement patterns of workers. That data stream feeds centralised dashboards. The official discourse is workplace safety. The operational reality is the synchronisation of the biological body with the algorithmically determined pace of production. The worker no longer contributes merely his time and his gesture. He contributes his vital signs. And his vital signs, like Amazon’s TOT, are recorded without his having access to the logic that interprets them or to the decisions they produce.
The Algorithm as a Class Weapon
In 2021, workers at an Amazon warehouse in Bessemer, Alabama, voted on whether to unionise. A study published in the journal Socius in March 2025 by researcher Teke Wiggin of Northwestern University documents the specific role of algorithmic management in that process. The same devices that directed the work, assigned tasks, recorded performance and issued automatic warnings were used during the campaign to transmit anti-union messages on workstation screens. At mandatory meetings where the company presented its arguments against the union, supervisors verified in real time the disciplinary score of each worker present, making visible the algorithmic eye that would watch them when they left the room. The system that controls the work was converted into the instrument of the political campaign against those who perform that work. The algorithm’s neutrality was revealed, at that moment, as what it had always been. A cover narrative.
At the Bessemer, Alabama warehouse, workers voted against unionisation by 1,798 votes to 738, from a total of 5,867 eligible workers. According to Teke Wiggin’s study published in Socius in March 2025, algorithmic labour control devices were actively used by the company during the campaign to discourage union organisation.
Resistance and Its Glass Ceiling
Legal and union responses exist and are real. Spain adopted the riders’ law in 2021, obliging platforms to disclose their algorithmic parameters to workers’ representatives. Courts in the Netherlands, Italy, Spain and Colombia have recognised the employment relationship concealed beneath the collaboration contract. These advances share a common limit that none has been able to overcome. Algorithmic transparency in practice means receiving technical documentation that no worker, judge, or trade union has the means to audit independently. A rider contesting an account block may receive dozens of pages of system specifications describing, in the language of the system, the functioning of the system. What is delivered as transparency is another layer of opacity with better presentation.
The asymmetry is structural. The worker contributes his body, his time, his tools, and, in the most advanced cases, his vital signs. The company contributes the code. The code processes the body’s data and produces decisions about time and money. The worker observes the result without access to the logic that produced it. This is not resolved by accumulating laws that demand more documents. It would be resolved by the real capacity to audit, challenge, and modify the systems that determine working conditions. That capacity does not exist in any legal order currently in force with the depth the problem requires.
Impunity Without a Subject
What algorithmic management has achieved is not, at bottom, more exploitation than existed before. It is exploitation without an identifiable party responsible. When April Watson cannot obtain the correct medical form because Amazon’s human resources artificial intelligence system cannot find it, nobody made an error. Nobody took an incorrect decision. The system produced a result. When the Rappi rider is blocked without warning and without the possibility of immediate human recourse, nobody dismissed him. The algorithm registered an anomaly and acted accordingly. When the Bessemer workers voted against the union under the constant pressure of the digital eye counting their minutes of inactivity, nobody formally coerced them. The environment was managed algorithmically.
The history of industrial capitalism is, among other things, the history of the attribution of responsibility. The employer had a name, a face, signed the contracts and could be sued. The foreman was on the factory floor. The violence of extraction was real and had identifiable agents. What algorithmic management has produced, as the most mature form of work organisation, is the perfect dissociation between extraction and responsibility. The system extracts. Nobody is responsible for the system. This dissociation is not accidental. It is the most important political innovation of platform capitalism, more important than labour flexibility, than the reduction of fixed costs, than global scale. It is the construction of a relation of domination without a recognisable dominant subject, and therefore without an accusable subject. April Watson’s body is still without the form. The counter, for its part, never stopped…
G.S.
