The Death of the Operator

The term “automation” was coined at Ford Motor Company by Vice President of Manufacturing Delmar S. Harder in 1946. This coining coincided with a number of pivotal breakthroughs in computation that emerged from World War II–related technical innovations, including cryptography, electrified calculation, and ballistics. That year, women—referred to as “computers”—programmed ENIAC (Electronic Numerical Integrator and Computer), the first electronic, general-purpose, Turing-complete digital problem-solver at the University of Pennsylvania. In St. Louis, the first mobile phone service was inaugurated by Bell Labs, connecting an automobile’s engine and radio transmitter to telephony and thus enabling virtual communication on the move.

The school of “cybernetics” led by Norbert Wiener blossomed. Back in 1834, French philosopher of science André-Marie Ampère had used the term la cybernétique to refer to the art of governing, drawn from the Greek kybernetes, meaning “navigator” or “steersman.” Wiener’s cybernetics defined an interdisciplinary field of study investigating control and communication in mechanical and computational systems, including feedback, coordination, and management. The term “automation” seemed to encapsulate the new emphasis of programmable relationships for a world increasingly shifting from clocks to code.

At Ford Motors, the new automation department developed integrations for sensing and feedback applications into Ford products and processes. At Ford, automation was not framed as a new phenomenon but was rather understood as a critical step in the evolution of mechanization, following on from the famous electrified assembly line that enabled the mass production of Ford’s Model T from its inception in 1914. Fordist mechanization had been enabled by continuous breakthrough choreographies of stringent precision protocols—mass production was made possible by vertical and horizontal integration, standardized parts increasingly produced in-house with an ever-diminishing tolerance of differences, repetitive labor bolstered by high wages, and a steady resources flow sourced from global locations. This enabled a predictable calculation of time and resources that flowed into regimented production.

The term “automation” at Ford referred to the process of “transfer automation,” whereby various independent machine processes—pressing, sorting, cutting, pasting, welding, twisting, scanning—were linked by positioning devices that provided a continuous, unstaffed production flow between sub-constituent parts of assembly. Automation connected machine to machine without a human intermediary, eliminating the need for operators on the factory floor.

In 1954, as tensions about the future of automation and the impact on labor reached a political pitch, including congressional hearings, Ford’s automation department doubled down on the notion that despite being a new term, automation was an extension of a process that Ford had been pioneering all along. “We do not believe that automation, as we use it, is a revolutionary development in production technique; rather, it is just another evolutionary phase of our advancing production technology.”⁵ Henry Ford had described the power of mechanization without human oversight in his bestseller Today and Tomorrow, published in 1926. “In the Chicago plant, the greatest distance any material has to be trucked is twenty feet, this being the distance from the incoming freight car to the first conveyor. After this it is mechanically handled during the entire process of assembling it into a finished car.”⁶

The term “automation” was concurrently deployed by Harvard management consultant John Diebold in 1946, who describes how the oft-used term “automatization,” which was itself short for “automatism in production,” was a veritable “tongue twister.”⁷ Diverging from Ford’s prevailing argument of continuity, Diebold characterizes “automation” as a new paradigm distinguished by the potential to fully mechanize mechanical or computational production processes. For Diebold, automation was a phenomenon specific to an age in which machine independence became possible through programmed control and predetermined decisions. He argues that even though automation was rightly understood as an evolutionary extension of a long process of mechanization that had been developing since the advent of the wheel, it also marked a distinct breakthrough in machine coordination, digital sensing, and feedback in physical work. Notably, Diebold understood this evolution in terms of tools and technicity, which, it could easily be argued, go back much further than the wheel. “If automation means anything at all, it means something more than mechanization. It implies a basic change in our attitude and manner of performing work. It is a new way of organizing and analyzing production concerned with the production process as a system, and a consideration of each element as part of the system. It is something of a conceptual breakthrough as revolutionary as Henry Ford’s concept of the assembly line.”⁸

Diebold’s quip about said paradigm shift, “an age when buttons push themselves,”⁹ encapsulates all the vicissitudes of this purported epoch of automation and the limitations of iterative imagination. It was an age where buttons still existed.

Diebold described automation as a “manufacturing philosophy” that goes beyond being merely a technology. It demands, he insisted, a thorough analysis of the entire production process, from raw materials to the finished product, ensuring that each step contributes as efficiently as possible to the enterprise’s overall objectives. “Automation is not a particular group of new machines or devices. It is a new concept— the ideal of self-regulating systems— and a new set of principles.”¹⁰ Encyclopedia Britannica’s entry on “mass production,” ghostwritten by Henry Ford in the 1920s, points to this whole-systems approach in planning an operation down to the last detail. “Mass production is the focusing upon a manufacturing project of the principles of power, accuracy, economy, system, continuity, and speed.”¹¹ A precision choreography of the entire process was required to establish and maintain the steady operation of the assembly line in the first place, from the machining of precision parts to the procurement of resources to the high pay that kept people manning the line and ensured the distribution and sales of what it put out. Ford Motors knew from the outset that whether through advanced mechanization or automation, the organizational cascades required to orchestrate and integrate new technologies were not only feats of mechanical engineering. They included organizational, social, and political technologies that were engineered alongside innovations in hardware and software.

Management scholar James Bright performed extensive case studies of the impact of the new phenomenon of automation on organizations and management systems. He astutely describes automation as a technological phenomenon distinguished by advancements relative to whatever existing milieu of mechanization it evolved from.¹² According to Bright, automation referred to something significantly more mechanized or automatic than its previous iteration.

Ford of course hadn’t invented the moving assembly line for car production. He famously lifted the idea from a moving meatpacking disassembly line, which drew on an incredible lineage of linear and regulated production techniques—from Chinese pottery to shipbuilding to brickmaking. The real Fordist innovation was the integration of updated technologies such as electrification and iteration—perfecting a coordination and choreography of electrified mass production within an even broader global matrix of labor, material resources, standardized parts, and machines that made machines.

With Fordist automation, humans were fully relieved from the manual day-to-day work of operating the machines. People were still required to design, program, organize, arrange, clean, maintain, manage, oversee, and supervise the machines. The dawn of so-called automation marked the death of the operator on the factory floor. Now, over a century later, the same principle is being realized in the operation of the automobile itself, freeing the driver from pitifully glorious routine operations and those delusions of control that prevent humans from pursuing a higher purpose.

Platform Programming

The chassis of the Model T is the foundational base on which the wheels and drive shaft connect, a standardized platform on which the rest of the car was built. This was also the first car part to be automated. A. O. Smith & Co., working with Ford, had already figured out how to eliminate humans from the physical manufacturing process of the chassis by 1908. Platforms and automation evolved in tandem. General Motors (GM), later competing with Ford on the margins of styling and cultural variability, would instantiate what it referred to as a “platform strategy.” This meant that GM produced standardized chassis with a variable and customizable design to try to appeal to global customers. In this case, a highly standardized platform centralization enabled a highly variable and responsive decentralization of design. This wasn’t so different from the Model T, which also offered a range of styles and designs particular to its buyers, but all were built on the exact same base.

The analytical engine was the first modern computer, a machine running on steam that could perform calculations that previously only a human computer had done. Before designing it, Charles Babbage closely studied the factory and its division of labor. On the Economy of Machinery and Manufacturers deconstructs the entire British factory system: the process of procuring raw materials, the mode of organizing workers, the disappearance of middlemen, the allocation of prices, the geographic positioning of factories, the importance of scale and velocity in realizing overall productivity, the impact of wages, and the dynamics of management. One passage discusses the factory and the division of labor as advantageous to both “mental and mechanical operations,” referencing French Revolution timetables as the preliminary means of converting to the decimal system.¹³ What once required personal mathematics could eventually be referenced on an existing code sheet, which transcribed conversions with alacrity. Ascribing a time code to every interlocking moving part was critical not only for managing the factory gestalt but also for imagining everything as an element of computation. The technique of taking systems apart and rationally organizing them part by part, piece by piece, moment by moment, established platform automation as a core logic of computational design.

From Chaplin to Durkheim to David Harvey, critics over the subsequent decades deconstructed the impact of the grueling requirements of assembly time on the humans who worked the line day in and out and the broader impact of Fordism on almost every aspect of modern society—from fatigue to economy to psyche. The critiques often conflate Taylorism and Fordism, with these philosophies’ mutual emphasis on efficiency, timekeeping, and work that privileged the mechanical clock over care for the human subject. In fact, Ford was careful not to mention Frederick Taylor and saw him as more of a management charlatan than a smart organizational designer, notably commenting that Taylor was always refining processes that didn’t need to be there in the first place.

Rather than focus on the assembly line only via its reductionism of subjects, we might also see this shift in human responsibility as a painful hurdle in externalizing labor, shifting from humans as independent workers to humans as reproducers of machine proliferation. “Humans became appendages,” as McLuhan states, “the sex organs of the machine.”¹⁴ McLuhan follows Marx’s fragment on machines here, which positioned the factory as a platform poised to liberate humans from its immediate demands via machines that make machines as much as a place for ongoing entrapment into the vicious cycle of production.

The insights cut another way—the best way to design a fully automated system is to act it out: to perform it, program it, plan it, and coordinate it in advance. In hindsight, the assembly line appears as a theater for rehearsing the next steps of automation, a mechanism of performing the future. Just as the horse had pulled the carriage, ultimately leading to mechanical horsepower, the workers on the Fordist assembly line were playing at being machines.

Automation as a Planetary Process

Buckminster Fuller was a champion of automation, and he lauded Ford as an artist and design visionary. This was partly due to Ford’s focus on finding increasingly light and durable materials, but it was mostly due to his total reengineering of the production process and society alongside it, his demonstrated commitment to converting luxury to mass mobility, and the unfailing ambition and total systems purview of his vision. As Fuller wrote in Critical Path, “Through the invisible and ever-higher-performance-per-pound alloys and the invisible controls of ever-closer measuring of invisibly operating parts of the machinery, structure, and production tooling of his automobiles, Ford developed the use of moving assembly lines. He concerned himself directly as the prime designer not only of his end product—the automobile—but also of his evolving machinery and structural technology and all the other supporting activities of final pertinence to the success.”¹⁵

In the chapter “Automation” in Operating Manual for Spaceship Earth, Fuller describes “comprehensively commanded automation” as a fundamental principle for a viable future Earth: “Automation displaces the automatons,” Fuller claims, positioning automation as a foundational process of biological and technological evolution, a dynamic biotechnical process of encoding and embedding persistent decisions.¹⁶

Fuller’s framing of automation presents it as an evolutionary dynamic, in which one thing that was once open to decision resolves itself into a predetermined context, setting a particular path that in turn enables other emergent formations. Automation is a process by which decision-making occurs through inscription into devices or environments, causing further cascades. In this way, automation accounts for a deeper evolutionary logic whereby what once required explicit cognitive oversight or decision becomes embedded into infrastructures, becoming part of a general environment and integrating into a broader platform unconscious.¹⁷ If automation is marked by the consolidation and congealing of a decision into a predetermined process, it is in fact an operational principle of incorporating feedback such that it is driverless.

In this sense, automation is highly contingent and path-dependent, as it entrenches platforms upon platforms.¹⁸ Platforms establish a programmable and distributed mode of patterning intelligence, meaning that decisions can be encoded or embedded into the system. Automation is not something only assembly lines do but also something that ecosystems do, that cells do, that psyches do, that bodies do, and that planets do. Evolution resolves decisions into automated forms, discovering structural formations through encounters and tests in environments rather than programming these operations from first principles. In this sense, evolution itself is a process of automation, one that generates and builds on a persistent path so that a choice that once required conscious decision no longer requires that decision to be reviewed at a conscious level.¹⁹

Persistent Prospects

In the age of AI, this delegation of decisions from humans to machines operates less like programming and more like learning. Learning can be understood as a process of updating a model based on feedback, one that advances the foundational framework to build other things and decisions on. Under this guise, we can view infrastructure as an evolutionary demonstration of platform automation, the collective externalization and amplification of labor that shifts intrinsic functions to external, embedded ones. Platforms enable automation to coordinate across individual nodes in a network so they become interoperable and to transfer infrastructural knowledge from one domain to another.

Automation will continue to pervade the vast majority of spaces, from closed dark factories to urban landscapes humming with robotic agents, and now in more cognitive modes, from writing to editing to imagery. Just as censored doors, switchboard circuits, and automatic banking machines once revolutionized services, the next waves of automation will take both sensational and banal forms: horizontal escalators, mobile street sweepers, delivery bots, payment processing, and psychological soothing. In factories across Earth today, fenced-off robots and their limbs that pivot in every direction are already patiently enacting the procedure of precisely installing and bolting parts of the automobile along moving assemblies, one after the other after the other, a perfectly patterned production lullaby.

Automation once provided a distinguishing mirror of the incredible power of the general intellect and the breathtaking stupidity routined by the inflexibility of systems incapable of parsing the exception from the rule. It was once thought that the banality of tasks absorbed by the increasingly interoperable machinery of production integrated via its decentralization would free the depths and dimensions of human curiosity and acuity. Today, as AI takes on both cognitive and physical dexterities once preserved for the human intellect, both language and hand are bending toward other, more evolved forms of automation that operate at scopes and scales impossible for human decision to apprehend or human labor to address.

Extrinsic Platform Power

The broadest lineage of the automobile can be traced back to the cart, or a simple platform on wheels, but the first instance of the modern automobile is attributed to French inventor Nicolas-Joseph Cugnot who created a small, steam-powered cart in 1769. Unlike the carriage, pulled by a horse or a few horses, the auto-cart—which eventually became the motorcar—had the capacity for self-propulsion and contained its power-processing mechanism within its own bodily construction.

By the late nineteenth century, the term “automobile” had pervaded the common vernacular, referring to a variety of vehicles powered by steam, compressed air, or electric motors. Unlike the locomotive, which was constrained by tracks, the automobile implied autonomy from the restricted collective trajectory of the train, an individual directional independence that afforded a flexible accommodation to any sudden shift in destination. The engine of automotive production was propelled by the internal combustion engine. Ford’s Model T housed a “planetary” engine, a name that referred to the design structure of its rotation but also, more poetically, to its capacity for wandering.²⁰

The core paradox of the auto remained. While the automobile’s power processor was carried on board, affording a certain degree of mobile independence, its energetic resource came from elsewhere. This was not a bicycle, which relied on a closed or contained power loop, requiring only the energetic momentum of its rider to propel itself forward. Instead, in its energetic contingencies, the automobile shares a close lineage with the automata that preceded it, and the long discussed uncanny vicissitudes of dependence and independence invited by these mechanical creatures that appear to move on their own.

Automata long influenced the human imagination through delight and disquiet, little test cases of “life” manifesting scientific invention and the fantasy of movement without human intervention. Independence was a delusion, a false boundary of the housing of the power-processing mechanism as the boundary for “self” containment. Early automata, from clocks to figurines, used hydraulics and pneumatics to regulate, harness, and modulate wind and water power, demonstrating cybernetic principles long before the term was coined and circulated. Descartes wrote poetically about Vaucanson’s Mechanical Duck, and the Mechanical Turk and intricately engineered figurines of piano players, dancers, dolls, and other devices were fascinations of the early industrial age. The mechanisms of biological life came to be understood through the metaphor of the mechanical. “The development of automatons made explicit an ever-unfolding series of correspondences between the body and the machine.”²¹ And, as Descartes put it, “The mechanisms of the body are the same as those of mechanical waterworks.”²²

These automata, while seemingly independent, all relied on power generated by external outside energies, kinetic power, or movement generated by concealed human puppetry. When the famous Mechanical Turk was displayed in 1793, Karl Gottlieb von Windisch wrote a pamphlet called Inanimate Reason, in which he discusses two deceptions of automata—“via motrix” (motive power) and “via directrix” (directive power).²³ These dual deceptions described both the deceit by automata and the delusions of its observers, with the mechanism or motion converter contained inside reflecting a displacement from the actual (external) source of energetic power and a predetermined prescription in the direction or decision of automated movements. This deception of agential autonomy similarly reflects the deluded condition of humans, constrained by motive and directive both by resources and via intelligence.

Motive suggests purpose and direction suggests intent. Darwinism makes clear that evolution is more responsive to immediate conditions than to long-term plans, constraining motive. Jessica Riskin points out that self-volition is never truly independent but is always independent, contingent, and externally constructed.²⁴ The proclivity for mobility at the core of life, as that which moves and adapts and mutates, is traceable to cellular volition, planetary chemistry, or electric currents, whether via wind, bacteria, mitochondria, or DNA. Life itself is not typically a conscious choice or decision at the outset, and yet movement occurs through an action or drive, first toward survival but then in response to other factors, many of which remain undisclosed to the agent driven by drives.

Most decisions are determined not by conscious directive but by a momentum of life shaped by social forces unintelligible to the conscious mind and often without intelligent purpose. Freud famously disconnected drives from natural instincts, instead linking them to political and social libidinal constructs. Biologist and philosopher Étienne-Jules Marey reviewed the motility of everything that moved without intent at a microcosmic scale, defining complexity as a set of trajectories that rejected an overarching directive for grand systems, even when they culminated as such.²⁵ While humans were well understood to be governed by the automatic function of internal organs, directed, conscious volition was a rare exception.

In this sense, the force of life can be understood best as momentum rather than motive, a form of evolutionary automation whereby the continued persistence and perpetuation of processes is preordained. Via automobility, or the automaton that moves on its own, it was demonstrated that individual survival and continuity were markedly different from the directive or trajectory of a species at large. Thanks to the auto, any agent, following its own momentum, could unwittingly pursue a path counterproductive to its larger species.

Motion Pictures: Drive and Desires

Ford famously said, “You can have the Model T in any color as long as it’s black,” a public quip that generously pandered to his populist audience while reinforcing the values of utility. You could originally get the Model T in many colors, but the drying process was quicker with black, and so efficiency won over choice. You could, however, purchase the black Model T in many versions, such as the runabout, the touring, the roadster, the speedster, the touring car, the coupelet (an early convertible with a folding top), and the torpedo (a streamlined sporty edition). Fordist vertical integration meant that Ford had more and more machines that made machines. But the chassis, the foundational platform, was exactly the same in every single version.

However, engineering the production process of mass production was not enough; mass consumption had to be engineered as well. When Ford purportedly famously joked, “If I asked people what they wanted they would have said a faster horse,” he was referring to the inability of the public to peer into the future but also to the required design of demand and desire. The low price was an important part of the equation, but social desire was not driven by price alone.
It was no coincidence that the instantiation of the assembly line coincided with Ford Motion Pictures, the largest film company of any industrial producer at the time. Mass production was nothing without steady demand that perpetuated the continuous circulation of automobility machines. Ford’s social platforms included profit-sharing schemes that made its workers owners, Ford newspapers that promoted Ford’s populist perspectives, the Ford Motion Picture Unit proliferating newsreels that just happened to feature a jet-black Model T in the most contemporary sagas of the day, and Ford dealerships that created clear protocols for Fordist maintenance and distribution. Ford developed signature methods for naturalizing strange technologies, developing platforms for production and dissemination and automating the drive for the automobile.

Automata and Driver-Cars

Behind the wheel, humans merge with the car, becoming, as John Urry described, “a car-driver or a driver-car.”²⁶ Extending the human frame, the autobody of the automobile merged the human body and the machine, inextricably intertwining the physical and the psychological. The automobile demonstrated an anti-Cartesian modality, in which cognition and coordination manifest in the rote and pleasurable operations of the drive.

With the fragile frame of the human body encased in glass and steel, augmented by the windshield and four wheels, the organization of mobility around the individual driver reinforced the delusion of individual automobility, of each human at the center of control of their own little universe and each automobile an island, protected privately from the greater world. By the simple pressure of a foot on the gas pedal, hands on the wheel, key in the lock, mirrors extending eyes in many directions, the gasoline motorcar became a prosthetic that supplanted other paths for the wheel or the pedestrian, which may have had more energetically sensible or rational means. There were other paths not taken, mostly due to capital and circumstance, from electric to bicycle power.²⁷ This role, this delusion of autonomy, was reinforced in reality and its myriad renderings, from sublime landscape scenes to movie screens traversed at unprecedented speeds.

There was always something larger than life about the automobile, something that animated it and gave it a personage, always seeming on the precipice of becoming sentient. In Ford’s day, the Model T’s were affectionately referred to as the Tin Lizzy, with songs and slogans devoted to the popular car that was durable and finicky at once, reliable but requiring special techniques acquired by its owner to maintain smooth functioning. This proclivity for animation, perhaps a function of the motive and directive delusions, has perpetuated car narratives ever since. In sci-fi writer Isaac Asimov’s short story “Sally,” the horror wish of automobiles deciding their own destination is depicted.²⁸ In The Uncanny, Freud describes the automaton as a figure of both familiarity and alienation.²⁹ The uncanny was a functional effect of life-likeness, of too-close-for-comfort proximity between human thing and inanimate thing that appears animate. Masahiro Mori used this term in relation to contemporary robots, describing an “uncanny valley” in which the more closely the mechanical thing resembles a biological living thing, the more threatening or disturbing it is to the living thing.

The emerging paradigm is something different. As the automobile shifts away from a machine that could be managed, memorized, and driven by humans, another philosophical paradigm is emerging, one in which the robot is not a single agent but part of a swarm of partially autonomous automata. But rather than total synchronicity from top-down control, the network of automobile automata maneuvers through a hybrid of centralized and decentralized interactions. It’s not that the automobile mimics something recognizable, but that it hybridizes the cart and the horse, the phone, the computer. It is moving from machine to robot, a computer to ride in, a single instance of an entire mobile intelligence network traversing the city. The driver, with their delusions of destiny and destination, of control and power, is dead.

The automobile will be less a single robot with its own desires and determinations than part-ride, part-simulation, part-body, and part-computer. It is a node in a moving, distributed sensing apparatus that feeds and is fed by an endless stream of data. These data comprise the computer city, which is networked to the planetary processes of outputs and inputs. The cloud car is not a car at all, but rather a terminal, a port, a synapse, one form of an urban-scale sensing-machine learning megastructure. It processes data as much as it processes physical transit.³⁰ Rather than have a mind of its own, the automaton shares a body and a mind and is part of a mobile planetary appendage roving the Earth.

Presumed Autonomousness

The earliest provenance of AVs can be traced to a little cart designed by Leonardo da Vinci. The cart consisted of a platform on wheels powered by two symmetric springs that were calibrated by the same sort of balance wheel commonly used in clocks. The cart also had a little fork attached to an arm that extended past its front base, a sensor that indicated when to turn. The little mobile automaton entertained spectators by seemingly autonomously navigating a predetermined path through Florence’s streets, following a series of well-positioned stakes indicating turning points throughout the city.

Like most automata, Da Vinci’s cart was preprogrammed to perform sequences of operations without human intervention. But it wasn’t the route itself that was programmed, but the reflexive and responsive behavior of the cart, based on interactions with the environment. Telling the car how to operate was not about operating the vehicle but about enabling the propulsive power mechanism by placing sensing devices on the agent and indicators in the environment. The vehicle’s mechanism was instructed to recognize these indicators with a specified routine response to determine its navigation.

The most notable aspect of the story is that Florence’s streets were one-way, making it easy for the cart to follow motion and avoid crashes. If this was indeed the first AV, as is often indicated, it follows that autonomousness not only refers to a device operated without external control but also implies the ways that that autonomous agent is able to leverage or resist environmental constraints.

This primal scene of self-driving plucked from the Renaissance indicates the ongoing tension between the design of control within the device and the design of control over the environment itself, demarcating a 500-year confusion over the focused calibration of such effects that continues today. Autonomous cars have been described by some as the ultimate “AI” problem—indeed, the challenge is not only pattern detection and decision in a computerized brain but also that the autonomous car is a physicalized exoskeleton, an embodied computer that humans ride in, bringing questions of safety, interaction with the environment, and security to the fore. Cognition in the wild forces the question of constraint as a factor of autonomousness.

The Autonomous Platform Stack

The successful acceleration of the automotive system required the convergence of a tremendous number of systems—the various systems of the vehicle itself, manufacture and assembly and distribution, fuel infrastructure, road infrastructure, supply chains, regulations and governance, traffic management, insurance and financing, education and training, etc.—and schematized production schematized it. Today, autonomous cars are increasingly possible thanks to a similar convergence of computational capacity—sensors and high-fidelity scanners, data processing, GPS, 5G connectivity, lidar and radar, object detection, classification and tracking, planning and decision-making, longitudinal and lateral control, etc. This autonomous automobility stack itself pressurizes and responds to the larger automobility ecosystem. It is fraught with the promise of driverless vehicles and concerns over data governance, power, security, and liability. Each item’s shape also has implications for interoperability (of hardware and software between tech systems), for multiplicities (or the various ways in which a single data point may be used), for modularity (which changes the logics of the production line), for the evolution of machine vision and its evaluations of trust, and for the regulatory environment for testing, deployment, and international exchange. Each of these specific technologies inhabits a domain that has implications that stretch far beyond a single automotive machine. The successful or stunted development of these various elements has far more to do with shifts across the platform automobility stack than with agent-level decisions.

Always-Incomplete Autonomousness

In 2014, the Society of Automotive Engineers developed a standardized framework for comparing levels of vehicle automation. The six levels of automation provide a useful heuristic for understanding automation more broadly, an epistemological and ontological philosophy of machine operation and interaction.⁴⁶ At Level 0, there is a human driver with no automation. At Level 1, the machine provides driver assistance (like adaptive cruise control). At Level 2, there is partial automation (the car can steer, accelerate, break in some circumstances). At Level 3, there is conditional automation (the car can navigate most of the environment, but the driver supervises). At Level 4, there is high automation (the car can navigate some geographic circumstances without driver input or oversight). At Level 5, there is full automation (the car can navigate all circumstances better than a human driver). Each of these levels describes an integration that is partially subjective, partially contingent, and discretionary.

At the time of their creation, the levels fueled a hype cycle in which various companies claimed that the next level was right around the corner. As 2020 approached, each news cycle featured another major car supplier insisting that it would be the first to fully realize AVs. After mostly minor altercations that destabilized public faith and easy confidence in the dream that driverless vehicles would happen overnight, this hype eventually faded into a crescendo of competition for the impossible dream of integrating all the phenomena required to safely enable autonomousness. Recent rollouts have been more moderately publicized, since the public is by now fairly acclimated to the eventuality of automobiles operating themselves. Sooner or later, any hype wears down into a new normal.

For scientific research, the levels of autonomousness sometimes aided and other times confounded the conversation around AV development, providing a regulatory framework for technical and policy response while often distracting researchers with the goal of designation levels rather than focusing on the nuances of the specific issues at play. Designating the levels of independence diverted researchers from the more difficult and prescient questions about the fundamental interdependence between drivers, riders, and environments, an issue that is more about interaction, feedback, and governance dynamics in automated programming than about approaching automation and its total achievements as a discrete matter of fact.

Rather than realizing full automation through the cautionary world of perfecting things in simulation environments, Tesla, taking many insights from the Ford playbook, has gained market control by testing driving in the wild, infiltrating the ecosystem rather than beginning with a grand systems overhaul. Tesla launched with a luxury EV, then offering a cheaper and cheaper consumer vehicle. In early 2025, Tesla announced a pivot from planning an even cheaper consumer vehicle to an all-out robotaxi platform, sidestepping ownership challenges in favor of subscriptions that invite current owners to become platform shareholders. This pivot builds on Tesla’s fundamental platform logic, with the car increasingly a computer on wheels, requiring updates and maintenance to its software and hardware and now matched by a distributed platform focused on users rather than drivers. Certain ideas—such as a real prototype tunnel that looks just like a train subway built exclusively for Tesla cars—make it clear that artificial stupidity can be just as easily programmed and developed as artificial intelligence, entrenching our biases (in terms of not only the racial ones everyone seems to be exclusively concerned with but also our ideological instincts of finding and creating patterns overall).

The platformization of intelligent automation implies that the thing formally understood to be “a” car is not actually a robot independent of all other robots moving around the city. The car is a platform node in a much broader integrated platform of cars, part of the city platform in which transport is an integrated function. Sam Hind describes the car as a media machine, presenting an alternative platform stack of automobility that might include the following dimensions: (1) the (hardware) “skateboard” platform; (2) the (mobile) sensing platform; (3) the (modular) app/product development platform; (4) the (developmental) partnership platform, which includes corporate connectors; (5) the (connective) platform ecosystem, which includes all IoT and other software infrastructures; and (6) the (embedded) urban platform.⁴⁷ The ever-increasing convergence, layering, and interoperability between the computational platforms that govern urban agents and urban infrastructure form an urban platform stack, in which the transport layer composes an automobility network that services movement at large.

Through this process, the automobile becomes a sensing platform, a delivery platform, a moving sensor that not only senses for its own navigation but can also feed sensing for other platforms into a vast network. This means that the streets can and must be reprogrammed, too, no longer subject to the ridiculous autopoietic logic that expanded highways through iterative accommodation. The effort to shape the urban environment in response to the impending autonomous automobile platform rather than have the automobile platform fully succumb to the inanity built into modern automotive infrastructure might be most poised in places such as London, which, modeled after the horse and cart, never accommodated the car comfortably to begin with. Coordinated vehicle operations and fleet services inevitably mean a change in the city-computer overall, from disappearing surface hardware—lights, road furniture, the clutter of cars—to the opening of space for people, foliage, and other mobility modes. The sooner cities start to reorganize their interdependence, coordinating themselves around autonomous automobility, the sooner autonomous automobility will come.

As evolutions in AVs play out, a further reduction in the functions of driving threatens their full disappearance. Urry’s term “driver-operators” describes the human driver that both drives the vehicle (as a function of steerage) and manages some vehicle operations (supervising and maintaining some of its basic functions). A driver might no longer refer to an operator; they might be the director or designer of the journey in the first place, introducing high-level parameters by which the car-operator might take decisions, with some degree of selection in the factors that determine such and such route. To drive no longer means to control at the level of the immediate decision or operation but to power, propel, and compel at a deeper and higher systems level.

Notably, this degree of connectivity and contingency, the platformization of the automotive network in conjunction with the platformization of the city, may in many ways make the car overall less autonomous from its context, since each agent becomes algorithmically tied to all the other information and all other agents as it moves around the city and the city moves around it.

From Devices to Environments

Highways have always been machine spaces, “human exclusion zones” where an exoskeleton is required to navigate their affordances. Today on a highway, it’s quite easy for any motorized vehicle to reach full autonomous mode. It is harder to ensure that an AV is ready to navigate the unexpected worlds of urban streets, which have to share the dominant machine space with other mobile beings—pedestrians, dogs, delivery robots, skateboarders, etc.

Humans are almost always somewhere in the loop of the process of things becoming automated. It’s common knowledge for AV developers that it’s far harder to provide safety for Levels 3 and 4 than for Level 5, in large part because human attention does not choose wisely when it comes to the split second. The incessant information platform of the phone has become an extraordinary distraction for drivers, but it’s even harder to concentrate on supervising a machine, occupying only part of human attention, than on totally driving it yourself.

One of the first AV accidents took place in 2018, when a self-driving Uber in Arizona crashed into Elaine Herzberg, who was killed crossing the street with her bike. The incident occurred because the supervisor, who was meant to be overseeing the vehicle, was watching livestream and did not have time to override the failed detection system in the darkness. It is more notable that Elaine’s family did not sue the driver or Uber but the city itself, because the road affordances suggested an opportunity for safe crossing but did not actually provide a reasonable place to do so for pedestrians or cyclists like Elaine. AVs only exposed the limited design of Tempe’s urban infrastructure and its failures to accommodate multimodality in succumbing to a singular automotive logic.

For now, in America, China, Brazil, and many other parts of the world, the streets are a wilderness of hybrid machines that increasingly drive on their own, intermingled with distracted humans trying to operate them manually. Sometime soon it will be difficult to fathom the ridiculousness of organizing urban environments around parked automotive machines. One can imagine an urban environment in which humans, or motorcars, or bikes, or golf carts, or buses might become the prioritized transport means to which the rest of the city bends. Driving games might become labeling games, in which humans teach AVs how to interpret and “read” the city and its various objects.

The AV problem is not primarily a device-or agent programming problem but mostly an environment-development problem. As we have seen, the more intelligent cultivation of a multipurpose environment would actually set the conditions and terms for AV development and the conversion of contemporary outmoded infrastructure, working backward from function rather than attempting to navigate the terrain accidentally developed at the behest of an automotive ownership principle based on delusional autonomousness instead of collective contingency.