The Repeatable Performer

I. The question of the week

The question can be put plainly, though the answers this year could not: when a system generates a voice, a face, a song, or a moving image on command, what becomes of the people who used to be paid to supply them? The occasions that raised it were specific — synthetic presenters that recite a script in dozens of languages, music platforms whose payout terms became contested, video tools launched to a chorus of studio anxiety — but the argument underneath is older than any of them. It concerns the relation between creative labor and the machines that reproduce its output, and it has a long lineage.

Nearly sixty years ago, in Understanding Media: The Extensions of Man, Marshall McLuhan wrote that “the message of the print and of typography is primarily that of repeatability. With typography, the principle of movable type introduced the means of mechanizing any handicraft by the process of segmenting and fragmenting an integral action.” That is the whole of this week’s topic compressed into two sentences. Generative media is the newest engine of repeatability, and its method is exactly McLuhan’s — it segments the integral act of a performance into promptable fragments and mechanizes the handicraft.

The arc I want to trace runs in two channels that did not stay parallel. The rhetoric began in a register of near-total optimism, peaked in the first quarter of 2025 in the language of democratization and innovation, and then turned — audibly — toward doubt. The reality moved on a slower and more ambivalent schedule: rapid adoption, uneven results, augmentation more often than replacement, and a persistent silence around the workers whose labor the technology most directly reproduces. Where the two channels meet is instructive. Where they miss is more instructive still.

II. What we’ve been saying

The vocabulary of late 2024 was the vocabulary of transformation, and it was almost entirely a workplace vocabulary. The Financial Times, in AI is transforming the world of work, are we ready for it?, framed the technology as a force set to “reshape the workplace,” with business leaders “eager to adopt AI tools to increase the productivity and efficiency of their workforces.” The Harvard Business Review counseled organizations on how to move from incremental to transformative use in How to Create Value Systematically with Gen AI, and the World Economic Forum surveyed early adopters on implementation in AI at work: A practical guide to implementing and scaling new tools. The register was managerial, confident, and almost wholly untroubled by the person whose craft was being automated.

One word did the heaviest lifting: democratization. Microsoft’s research arm titled a paper AI and the democratization of knowledge work, and the term migrated effortlessly from knowledge work to creative work, where it carried a particular charge. The claim was that tools once reserved for professionals would now belong to everyone. This was not a new promise. The Encyclopedia of Communication Theory records an earlier version of it, in which a networked medium would break “the severe technical constraints of the broadcast model, enabling a system of multiple producers, distributors, and consumers.” The metaliteracy literature had already announced the arrival in MetaLiteracy: “We are living in a time of immense creativity, with new opportunities for creators appearing nearly every day.” Generative media inherited that hope and rebranded it.

The optimism crested in the first quarter of 2025. A senior LinkedIn executive predicted that generative AI would “democratize innovation” and usher in an “innovation economy,” reported in Why a Senior LinkedIn Exec Believes AI Will Spark a Shift to an ‘Innovation Economy’. Marketers drew up their wishlists in The Ultimate Marketer Generative AI Wishlist. The European Broadcasting Union described the technology’s potential to “democratize content creation” in Generative AI: promising for content creation but not yet ready for primetime — though even that headline carried a qualifier that most did not. The dominant frame was time saved: From Time-Saving to Decision-Making: What’s Generative AI’s Next Leap for 2025? cited nine percent time savings on coding and three hours saved on a single piece of content, the sort of numbers that make the case for a tool without ever asking who absorbs the difference.

The first genuinely reflective voice arrived in the same quarter, and it came from outside the promotional apparatus. The paper Generative AI and Creative Work: Narratives, Values, and Impacts observed that the very concept of creative work was being “influenced by discourses originating from technological stakeholders and mainstream media” — which is to say the people selling the tools were also writing the story of what the tools meant. That is the sentence at which the rhetoric began, quietly, to interrogate itself.

By the second quarter the interrogation was audible even inside the boosterish sources. The Harvard Business Review published a study of more than three thousand workers under the unambiguous headline Research: Gen AI Makes People More Productive—and Less Motivated, conceding a cost that the productivity literature had spent a year eliding. This is the inversion point of the arc: not a reversal from praise to condemnation, but the moment the promise acquired an asterisk it could no longer shed.

The third quarter settled into a chastened maturity. A study of artists in The Rise of “Alone Teamwork” described the tools “redefining collaborative human-AI” creation rather than abolishing the human. Gallup offered reassurance in AI Is Changing Creative Work, but the Arts Aren’t Disappearing, a headline that concedes the anxiety it means to calm. And the enterprise press turned from evangelism to project management, as in How to Make Enterprise Gen AI Work and Beyond Content Creation: How Generative AI is Redefining Business Workflows. The word democratization had quietly retired; the word governance had taken its shift.

III. What’s been happening

Beneath the rhetoric, adoption moved faster than any of it. A study from MIT’s Center for Information Systems Research, summarized in AI Brew News, documented the rapid spread of “Bring Your Own AI” — workers importing tools their employers had neither sanctioned nor understood. The technology entered the workplace through the back door, ahead of policy, ahead of contracts, ahead of anyone’s decision about who owned what it produced. This is the first fact of the reality channel: usage outran the terms of usage.

The second fact is that the results were mixed, and honestly reported as such by observers with no incentive to undersell. S&P Global’s AI experiences rapid adoption, but with mixed outcomes found that generative AI adoption “outstrips longer-standing forms of AI, as well as enterprise forecasts,” while “project failure rates appear to be elevated, as organizations attempt to deliver generative AI projects at pace.” The tools worked and did not work at once — a pattern the EBU’s own pilot confirmed when Generative AI: promising for content creation but not yet ready for primetime reported that an animated production built with the technology “retains some limitations for now.” The frontier was real; it was also closer than the marketing implied.

The third fact concerns the creative worker specifically, and here the evidence points in a direction the displacement panic did not anticipate. Gallup’s AI Is Changing Creative Work, but the Arts Aren’t Disappearing argued that the assumption of mass obsolescence — “if software can produce images, music, and text in seconds” the artists “will be among the first” to go — was not borne out. What the tools did instead was restructure the act of making. The Rise of “Alone Teamwork” captured this precisely in its title: artists using generative tools now work as if collaborating with an absent partner, a solitude that behaves like a team. The integral action of creation was being segmented and fragmented — McLuhan’s mechanization of the handicraft, arriving on schedule, only now the handicraft was the performance itself.

The fourth fact is the one the productivity literature almost missed: the human cost is not primarily a headcount. The three-thousand-worker study in Research: Gen AI Makes People More Productive—and Less Motivated located the damage not in jobs eliminated but in motivation drained — the flattening of the felt experience of work even where the work remained. A performer whose voice is cloned is not fired; a performer whose voice is cloned is repeated, indefinitely, at zero marginal cost, while the original attends to something else. That is a different injury than unemployment, and the discourse of “the arts aren’t disappearing” is not equipped to see it.

The fifth fact is a silence, and it is the most revealing item in the record. The documentary corpus of this arc is overwhelmingly a corpus of vendors and consultants — Microsoft, Deloitte’s Generative AI for Enterprises, ServiceNow’s explainer What is generative AI?, the Harvard Business Review, the World Economic Forum. Even the sober voices are institutional; the university library guide Issues and Benefits of Using Generative AI weighs “enhanced productivity” against “pressing challenges” from the standpoint of the researcher, not the session musician or the dubbing actor. The people whose labor is most directly reproduced by these systems are the people least represented in the archive that describes them. That absence is not neutral. It is the shape of who gets to narrate a technology, and it repeats a pattern our own briefing on AI literacy of 2025-08-31 traced in the language of “democratizing access” — a promise that consistently issues from the parties positioned to profit if it is believed.

IV. Where they meet, where they miss

The two channels meet on one genuine point of agreement: creative output has multiplied, and more people are making more things than before. The metaliteracy literature’s forecast in MetaLiteracy of “immense creativity, with new opportunities for creators appearing nearly every day” is not false. The volume is real, and the Encyclopedia of Communication Theory’s old dream of “multiple producers, distributors, and consumers” has, in the crudest quantitative sense, arrived. On the question of whether the machine can produce media at scale, rhetoric and reality shake hands.

They miss on the question that actually matters, which is not can it produce but who is paid for the repetition. The rhetoric of democratization performed a quiet substitution: it described a redistribution of tools as though it were a redistribution of income, when the evidence shows these are not the same event. A performer can be more “empowered” — handed more capacity, more reach, more speed — and simultaneously poorer, because the value of the capacity has migrated to whoever owns the model that repeats it. This is the substitution that the word democratization was doing work to conceal, and it is why I distrust the word. Skepticism of that vocabulary is not cynicism; it is reading the sentence for what it actually claims.

McLuhan’s insight cuts cleaner than the boosters’ because it never confused reproduction with liberation. Movable type did democratize reading; it also professionalized publishing and impoverished the scribe. The mechanization of a handicraft always redistributes the handicraft’s rewards, and it rarely redistributes them toward the hand. The generative case follows the pattern with unusual fidelity. The arxiv authors saw it early in Generative AI and Creative Work: the story of what creative work is was being written by the “technological stakeholders” who profit from a particular answer. When the storyteller and the beneficiary are the same party, the story is evidence about the party, not about the work.

The reassurance genre — “the arts aren’t disappearing” — misses in a subtler way. It answers a question about extinction while the actual harm is about terms. The HBR productivity study already showed that the injury of these tools registers as demotivation, dispossession, the draining of authorship — none of which shows up in an employment count. As our briefing on social aspects of 2025-08-10 argued, the systemic questions — transparency about training data, accountability for whose work was ingested, the terms on which a likeness is licensed — are precisely the ones the optimistic frame is built not to ask. The arts survive. The artist’s leverage is what erodes, and leverage does not appear in a headline that only knows how to count jobs.

V. The longer view

The pattern in this arc is not that a hopeful story met a grim reality. It is that a hopeful story, told mostly by the parties it enriched, met a reality that was more ambivalent and more quietly consequential than either euphoria or panic could register. The performer did not vanish; the performer was repeated, and the difference between being replaced and being repeated is the whole of what the discourse failed to name. McLuhan understood in Understanding Media: The Extensions of Man that every technology of repeatability mechanizes a handicraft by fragmenting an integral act — and that the reward for the act tends to leave with the mechanization. Generative media has extended that logic from the printed page to the human face and the singing voice, and the people who supply those faces and voices are, tellingly, the ones our archive of celebration forgot to interview.