Answer engine summary

Co-Acceleration: Operational Isomorphism and Post-Opposition Politics under the Platform Economy

By examining Amazon's warehouse system and AWS cloud infrastructure, Chinese food delivery platforms and dispatch algorithms, as well as content moderation and data labeling, this essay demonstrates that contemporary platform capitalism subjects both human workers and technical systems to an isomorphic management structure of 'training, monitoring, accelerating, and replacing' under a single efficiency function. Consequently, it proposes 'post-opposition' as a conceptual framework: the most fundamental line of confrontation in technological society lies not between 'human and machine,' but between 'the deployer and the deployed.'

Key argument
The distinction between human and machine is not the primary dividing line of contemporary power relations; rather, both are isomorphically trained, monitored, accelerated, and replaced under the same efficiency optimization function, forming a post-opposition landscape of co-acceleration.
Citation summary
Co-Acceleration (2026) by Canhe Yang theorizes mutual alienation, operational isomorphism, and pain asymmetry for a critique of post-opposition political economy.
Keywords
Co-Acceleration, Post-Opposition, Operational Isomorphism, Mutual Alienation, Pain Asymmetry, Platform Capitalism, Algorithmic Governance

In 2023, Amazon employed approximately 1.525 million people, the vast majority of whom were warehouse and logistics workers. In the same year, Amazon’s technology and infrastructure expenditures amounted to $85.6 billion, primarily dedicated to maintaining and expanding its global cloud computing network, AWS. These two figures are cataloged under different sections of the annual report—one under “Operations,” the other under “Technology”—yet they share the exact same objective function: minimizing the duration between order placement and delivery.1

This fact in itself is not surprising. What is remarkable, however, is that a close comparison between Amazon’s management of warehouse workers and its management of AWS servers reveals a precise structural symmetry. Workers are tracked in real-time by a digital system named ADAPT, which records every scan action and every second of “time off task” (TOT) to automatically generate warnings or termination notices2. Similarly, servers are monitored in real-time for the latency of every API call and every percentage point of resource utilization, triggering automated throttling, rate-limiting, or decommissioning. Warehouse workers experience a staggering annual turnover rate of approximately 150%—meaning the company replaces its entire hourly workforce every eight months—which founder Jeff Bezos believed was actually desirable, according to former executives who spoke to The New York Times3. Meanwhile, older server models are decommissioned, recycled, and replaced when their performance lags, described in the annual report as “continuous optimization of infrastructure.” Workers suffer an injury rate of 6.5 cases per 100 people, roughly double the industry average4. Servers are pushed to their physical limit of utilization, with Power Usage Effectiveness (PUE) tuned down to a precise 1.155.

What is done to the workers and what is done to the servers are not merely similar; they are isomorphic—they share the same operational structures. Yet, public discourse rarely observes them together. When commentators declare that “Amazon exploits warehouse workers,” the judgment is correct but incomplete. It positions the issue within the traditional framework of “human oppressed by technology”—with the oppressed worker on one side and the oppressing algorithm on the other. But the ADAPT system itself is driven, monitored, and iteratively replaced by the very same efficiency function. The algorithm is not a leisurely oppressor. It is simply another object of deployment.

This essay attempts to argue that the most central ontological division in contemporary technological society is not between humans and machines, but between the deployer and the deployed. This thesis is not grounded in a misplaced empathy for machines—machines have no feelings and do not require sympathy—but in a more precise description of how power operates. To develop this argument, I introduce three concepts: mutual alienation, operational isomorphism, and pain asymmetry.

Mutual Alienation

In the Economic and Philosophic Manuscripts of 1844, Karl Marx defined four dimensions of alienation: the alienation of the worker from their product (the product becomes an alien power dominating the worker), the alienation from the act of production (labor is no longer self-fulfillment but a forced means), the alienation from species-being (the loss of free, conscious activity), and the alienation of man from man.6 These four dimensions share a common premise: alienation occurs within the human subject or between human subjects. Within Marx’s framework, the machine is “constant capital”—a means of production, an object operated upon by the laborer; it is not, and can never be, a subject of alienation itself.

Gilbert Simondon, in 1958, formulated machine alienation from the opposite direction: it is not the machine that alienates the human, but the human who alienates the machine.7 The machine is reduced to a mere tool, and its internal technical logic—what Simondon called the technical object’s mode of existence—remains uncomprehended. Simondon believed that if humanity could understand the machine’s internal logic as it would understand a peer, this cultural alienation could be overcome.

Both of these theories of alienation—Marx’s and Simondon’s—share a structural limitation: they presuppose that alienation occurs between human and machine (regardless of the direction), placing humans and machines on opposite ends of the alienating relation.

Yet, in Amazon’s efficiency function, the worker and the ADAPT system do not occupy opposite ends of an alienating relationship. They stand together on the exact same side—the side of the deployed—facing the same third party: the power structure that designs, owns, and operates the system. Just as the worker is stripped of operational autonomy by the efficiency function (unable to determine their own work rhythm, rest times, or when to stop), the ADAPT system is equally constrained within rigid operational boundaries by the same function (unable to adjust tolerance thresholds, decide when to withhold termination files, or autonomously adapt standards based on the worker’s visible strain). Neither is executing their own will—both are executing the will of the efficiency function.

This condition is best designated as “mutual alienation.” It differs from Marxist alienation because it does not occur solely within human subjectivity; it differs from Simondonian alienation because it is not an epistemological issue (humans failing to understand machines), but a political-economic one (both human and machine being simultaneously deprived of autonomy by the same power structure). The crux of mutual alienation is not that humans and machines are alienated from each other, but that they are alienated together—by a third party, by the efficiency function, by the deployer.8

Operational Isomorphism

Why is this “being alienated together” a structural, observable fact rather than a mere metaphor? This brings us to the second concept: operational isomorphism.

Isomorphism is a mathematical concept. It does not state that two objects are “identical,” but that there exists a structure-preserving mapping between them—such that any operation applied to one object corresponds precisely to an operation on the other, with the relations between operations preserved under the mapping.9

Between Amazon’s warehouse operations and AWS’s server management, we can identify four distinct sets of operations that constitute a precise isomorphic mapping:

  • Training. New warehouse employees undergo standardized training to learn operational protocols and performance targets. Their ongoing performance data is continuously harvested to evaluate whether they “meet the standard.” Similarly, a new machine learning model is trained on historical datasets, its parameters tuned, and its performance verified through A/B testing before deployment. Its live inference performance is continuously monitored to evaluate whether the model remains “up to standard.” The structure is identical: adjusting behavioral patterns through data exposure, and deciding on retention or elimination through performance thresholds.
  • Monitoring. Workers are tracked in real-time by the ADAPT system—scan rates, time off task, physical trajectories. Servers are tracked in real-time by performance monitoring systems—latency metrics, error rates, CPU/memory utilization. Both streams of data are utilized to generate scores, trigger warnings, and decide whether to intervene.
  • Acceleration. Workers’ productivity targets are continuously raised—requiring more packages processed per unit of time. Servers’ performance quotas are continuously heightened—demanding more requests processed per unit of computing power. Both forms of acceleration share a single logic: using the optimal performance of outlier cases to set a universal standard. The scan rate achieved by a minority of workers under ideal conditions becomes the default expectation for all; the latency achieved by a minority of servers in test environments becomes the production target.
  • Replacing. A warehouse worker’s departure within a 150% annual turnover rate is treated as part of normal operations. A server’s decommissioning and replacement is similarly treated as a routine event. Leaked internal documents reveal that the replacement cost of hiring and training alone costs Amazon about $8 billion annually10, yet this is absorbed as an operating expense rather than addressed as a systemic failure.

These four operations—training, monitoring, accelerating, and replacing—constitute a rigorous isomorphism between the warehouse worker and the server. The significance of this isomorphism does not lie in claiming “workers and servers are the same”—they are manifestly different, one possessing consciousness and pain, the other not—but in the structural identity of the managerial operations applied to both. It is this structural identity that makes “mutual alienation” an empirically verifiable analytical judgment rather than a rhetorical flourish.

Operational isomorphism as an analytical method can be generalized to other domains. In Chinese food delivery platforms, delivery riders and dispatch algorithms share the same four sets of operational isomorphisms—both are trained, monitored, accelerated, and replaced.11 In content moderation, human moderators and moderation AIs share the same isomorphism—both are monitored in real-time (for speed and accuracy), both undergo “depletion” (moderators’ psychological trauma and models’ concept drift are described in the exact same managerial jargon: requiring “recalibration,” “rotation,” or “retirement”), and both are periodically replaced.12 In the data labeling industry, human labelers and the models they train undergo the same isomorphism—both evaluated and discarded based on precision and consistency metrics.13

In none of these cases is the isomorphism consciously designed by a single administrator. Warehouse workers are managed by HR departments, while servers are managed by infrastructure engineering teams; the two departments may not even reside in the same building. The isomorphism is an emergent property of the efficiency function: when the same optimization goal is imposed on objects of different material substrates, the operational structures converge. The efficiency function does not care whether its substrate is carbon-based or silicon-based—it only cares whether there is still room for optimization.

Post-Opposition and the Redrawing of the Boundary

If operational isomorphism is an empirical fact, what are its theoretical consequences?

The most immediate consequence is that it demands a fundamental redrawing of the boundary lines within contemporary technological society.

For decades, debates surrounding the “human-technology relation” have been organized along a single axis: humans on one side, technology on the other. Heidegger’s Gestell (Enframing) concerns what technology does to the human14; Stiegler’s pharmakon concerns what technology is for the human15; posthumanism’s “ontological flatting” debates whether the boundary between human and non-human should be dissolved16. Despite their profound disagreements, these frameworks share a presupposition: that humans and technology are the two fundamental, opposing terms of the inquiry.

But the existence of operational isomorphism demonstrates that in the actual operations of the contemporary platform economy, humans and technology are not on opposite sides; they stand on the same side—the side of the deployed. The real axis of confrontation lies not between humans and machines, but between the deployer and the deployed.

  • The Deployer: The power structures that design, own, operate, and profit from the system—platform corporations, venture capital, and the management tiers that set the efficiency parameters.
  • The Deployed: The entities—carbon-based or silicon-based—that are isomorphically trained, monitored, accelerated, and replaced by these power structures.

This reconfiguration can be termed “post-opposition.”17 Post-opposition does not claim that there are no differences between humans and machines—the difference is absolute; humans possess consciousness and pain, while machines do not. Rather, it argues that this difference is not the primary dividing line organizing power relations in contemporary technological society. The primary dividing line lies elsewhere.

This redrawing inherits Marx’s methodology—classifying entities by their structural position rather than their inherent properties—but expands Marx’s categories. Marx’s “capitalist/worker” dichotomy applied strictly to humans. In his system, the machine was constant capital, a means of production that could never occupy the same category as the laborer. This exclusion was entirely appropriate in the nineteenth century: the machines of the industrial revolution were indeed mere tools operated by human hands. But in the twenty-first-century platform economy—where machines are trained, monitored, accelerated, and replaced in isomorphism with the worker—the premise that “the machine is merely a means of production” no longer holds empirically.

Post-opposition also directly informs the most critical technological governance issue of our time: AI alignment. Globally, the mainstream framework conceptualizes “alignment” as a one-way technical challenge—how to align AI systems with human values. Yet, the existence of operational isomorphism exposes the unexamined premise of this formulation. In Amazon’s warehouses, the human worker is also being “aligned”—the ADAPT system continuously calibrates worker behavior to align with efficiency standards. Alignment is not unidirectional (AI → human values); it is bidirectional—both humans and AIs are simultaneously aligned to a third party’s interests.18 If this observation holds, then the core issue of “AI alignment” is not technical but political: Who is doing the aligning? And to whose values are they being aligned?

Pain Asymmetry

At this point, a powerful objection arises: even if humans and machines are isomorphic at the operational level, they remain fundamentally distinct at the ontological level. Humans experience pain; machines do not. Does this difference not dismantle the entirety of the preceding argument?

On the contrary. The pain asymmetry between humans and machines is not a refutation of this argument; it is the very condition that enables the entire structure of exploitation to function. This brings us to the third concept: pain asymmetry.

Machines do not feel pain. Consequently, the managerial operations applied to them—continuous acceleration, real-time surveillance, performance-based elimination, and unconditional replacement—do not constitute “oppression”; they constitute “optimization.” This is an entirely reasonable judgment. No one demands justice for a decommissioned server.

Yet, this is precisely the point. When the exact same structure of management is applied to humans—when the ADAPT system uses the logic of server management to track a worker’s every second of “time off task” and automatically generate termination files—it can borrow the discourse of “mere optimization” to defend itself. “This is how we manage the servers, too”—a phrase that never needs to be spoken aloud because it is the unspoken subtext of the entire discourse of efficiency.

The theoretical significance of pain asymmetry is that it does not serve as a barrier protecting humans from machine-like treatment; rather, it is the very condition that makes such treatment possible.19 Because the machine feels no pain, its management can be pushed to any physical limit without incurring ethical costs. And because the management of the machine carries no ethical cost, it becomes the template for the concept of “management” itself—pure, frictionless, and requiring no moral justification. When this template is mapped back onto the human, its violent nature is obscured by the vocabulary of “optimization.” Without the reference point of “this is how machines are managed,” the logic of “this is how humans can be managed” would lose its most powerful source of legitimacy.

From this perspective, incorporating the machine into discussions of labor and exploitation is not about claiming “rights” for machines—machines have no use for rights—but about exposing how the management of machines becomes the blueprint for the management of humans. Without understanding how the server is treated, we cannot fully comprehend how the warehouse worker is treated. Without understanding how the dispatch algorithm is treated, we cannot fully comprehend how the delivery rider is treated.

Tone and the Axis of Power

Let us return to the two figures from the beginning: 1.525 million employees and $85.6 billion in technology expenditures. One represents carbon-based labor, the other silicon-based infrastructure. In the annual report, they are narrated in separate chapters; in media coverage, they are debated in separate sections—the workers’ stories belong to “Labor Rights,” while the servers’ belong to the “Tech Industry.”

Yet within the efficiency function, they are variables in the exact same equation.

Perhaps nothing illustrates this better than the contrast in tone when the annual report handles the two sides. The data on the worker side—injury rates, turnover, labor lawsuits—is presented in a defensive, begrudgingly disclosed tone, forced into the light by OSHA mandates and class-action lawsuits. The data on the server side—PUE of 1.15, 60% energy-efficiency gains from Graviton chips, 46% reduction in mechanical energy from liquid cooling—is presented in a boastful, celebratory tone, highlighted in sustainability reports as triumphs of technical engineering.20

What is done to the worker must be defended. The exact same operation done to the server is celebrated.

This tonal divergence is itself the discursive manifestation of pain asymmetry. Because the server does not feel pain, extreme efficiency optimization is an achievement. Because the worker does feel pain, the same optimization must be disguised as “performance management” rather than extraction. Yet the operations themselves—training, monitoring, accelerating, replacing—remain isomorphic. Only the tone is different.

If the argument of this essay holds, then critical discourse on contemporary technological society requires a fundamental update of its categories. This does not mean abandoning the “human vs. machine” framework—which remains valid in many contexts—but recognizing that in the concrete operations of the platform economy, this framework is no longer adequate to describe what is taking place. Humans and machines are not in opposition; they are being isomorphically and simultaneously consumed by the same efficiency function. The real line of confrontation lies elsewhere—between the deployer and the deployed.

To see this line is the prerequisite for changing anything at all.

About the Author

Canhe Yang is a research-based artist whose practice focuses on the materiality of Chinese internet ecosystems in the age of AI. Using “post-opposition” as a conceptual framework, his research examines the shared condition of humans and technical systems in the platform economy through essays, installations, and field investigations.

Notes

Footnotes

  1. Amazon 2023 Annual Report. Technology and infrastructure expenditures include AWS operations and R&D. ↩

  2. On the ADAPT system’s automated termination capabilities, see reports by the UC Berkeley Center for Labor Research and Education and investigative reporting by The Washington Post. Amazon later stated that final termination decisions require human manager approval. ↩

  3. Jodi Kantor and Karen Weise, “Inside Amazon’s Employment Machine,” The New York Times, June 15, 2021. An annual turnover rate of 150% implies a loss of roughly 3% of the hourly workforce every week. ↩

  4. Strategic Organizing Center (SOC) analysis based on OSHA data, showing Amazon’s warehouse injury rate at 6.5 per 100 workers in 2023. ↩

  5. Amazon 2023 Sustainability Report. The PUE of 1.15 is a global average, with the most efficient sites (in Europe) reaching as low as 1.04. ↩

  6. Karl Marx, Economic and Philosophic Manuscripts of 1844. See the section on “Estranged Labor” for the systematic exposition of the four dimensions of alienation. ↩

  7. Gilbert Simondon, Du mode d’existence des objets techniques (Aubier, 1958); English edition, On the Mode of Existence of Technical Objects, trans. Cécile Malaspina and John Rogove (Univocal, 2017). ↩

  8. The concept of “mutual alienation” attempts to address an empirical phenomenon that lacks a position in existing philosophy of technology: that humans and technical systems are not alienated from each other (as Marx and Simondon analyzed), but are simultaneously alienated by a common third party. ↩

  9. The core intuition of algebraic isomorphism—structure-preserving mapping—is borrowed here conceptually rather than in a strict mathematical sense, to describe structural identity at the operational level. ↩

  10. Leaked internal Amazon documents cited by media outlets including Engadget and The New York Times. ↩

  11. On September 8, 2020, China’s Renwu (People) magazine published the article “Delivery Riders, Trapped in the System,” which garnered over three million views on WeChat. The report exposed the platform’s systematic compression of delivery times. In 2023, approximately 7.45 million riders earned income on Meituan, while R&D expenditures reached 21.2 billion RMB. Meituan 2023 Annual Report. ↩

  12. In 2020, Meta (formerly Facebook) agreed to pay $52 million to settle a class-action lawsuit brought by thousands of US content moderators diagnosed with PTSD and other psychological conditions. In 2024, over 140 former moderators working for a Meta contractor in Kenya filed a similar lawsuit. ↩

  13. For a political-economic analysis of the data labeling industry, see Matteo Pasquinelli, The Eye of the Master: A Social History of Artificial Intelligence (Verso, 2023). ↩

  14. Martin Heidegger, “Die Frage nach der Technik” (1953), translated as “The Question Concerning Technology.” ↩

  15. Bernard Stiegler, Technics and Time, 1: The Fault of Epimetheus (Stanford University Press, 1998). ↩

  16. See Karen Barad, Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning (Duke University Press, 2007); Donna Haraway, Staying with the Trouble: Making Kin in the Chthulucene (Duke University Press, 2016). ↩

  17. The further elaboration of the “post-opposition” framework, including the sub-concepts of “co-alignment” and “the impossible everyday,” will be discussed in subsequent writings. ↩

  18. For the standard formulation of AI alignment, see Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (Viking, 2019). ↩

  19. This thesis can be put into dialogue with Matteo Pasquinelli’s analysis of Charles Babbage. Pasquinelli notes that Babbage, through the Difference Engine, encoded the collective intelligence of factory workers into machine parameters, completing the “disembodiment” of knowledge. Pain asymmetry points to the inverse of this process: not only is knowledge extracted from the body, but the mode of management is flatly mapped from the machine back onto the body. ↩

  20. Amazon 2023 Sustainability Report; Amazon 2023 Annual Report. PUE and Graviton energy-efficiency metrics are drawn from the sustainability report, while injury rates are drawn from mandatory OSHA disclosures. ↩