The Portable Aesthetic

Seven years ago I argued that technical craft would be overtaken by taste. The labor market is conceding the point.

As more composers continue to look beyond audio as the sole medium for artistic creation and as these audio production techniques become more available and more simplified for a general audience to use, the barrier to entry for using these techniques, and thus their value, will lower significantly... The artist will have to learn how to take techniques and principles from their own discipline and rapidly apply them to new mediums, sometimes working with these new mediums for the first time and possibly the last time.

Josh C. Simmons, Towards a Postmodal Movement for Computer Musicians, PhD dissertation, University of California, Irvine, 2019

The passage above was written by a graduate student in a department almost no one on Sand Hill Road has heard of. It was read, generously, by four professors. The claim it makes has since been repriced by the labor market to the tune of several trillion dollars, though the repricing is still being performed, and the performers do not agree on what, exactly, they are pricing in.

The Dissolving Profession

Begin with the professions that broke first. A stock photographer in 2019 could charge hundreds of dollars for a competent studio image of a cup of coffee; in 2026 Midjourney generates a better one in eleven seconds for a subscription fee divided across ten thousand users. A junior illustrator who spent four years in art school learning to render convincing hands now competes with a model that, until recently, could not render hands at all and, more importantly, never will not render them again. Copywriters who billed at a hundred dollars an hour for product descriptions have watched their pipeline route itself directly to GPT-class systems that produce the same output in fractions of a cent. Voice actors, many of whom recorded union-rate scratch tracks for a living, now find that ElevenLabs has trained on a decade of their publicly available reels, and the broadcaster's budget line for "voice talent" has been reallocated to cloud compute. Translators who worked for twenty years building subject-matter depth in patent law, pharmaceutical regulation, or financial disclosure are discovering that DeepL performs at ninety-five percent of their quality for a tenth of a percent of their price.

These transitions are done. The prices have already reset, and the argument over whether they should have is a different argument.

The second phase is further along than most of its participants realize. Paralegals are now competitive with frontier-model contract review systems only when the contracts are unusually complex, which is to say: in the narrow margin of work where the client is willing to pay for insurance. Radiologists are being hired against a hybrid workflow that assumes AI pre-screening is the default, and the junior ones are taking a cut commensurate with the volume the machines now handle. Junior financial analysts, whose function was to assemble the deck, have been discovered by their managing directors to be replaceable by any person on the team willing to type a competent prompt. Customer support, once the entry-level white-collar job of last resort, is being rebuilt as an escalation layer on top of an AI triage system, and the humans who remain are paid for handling the cases the system flagged as too hard, which is a smaller and shrinking portion of inbound volume.

Dario Amodei, the CEO of Anthropic and a person whose public statements about AI are not idle, said in May 2025 that AI could eliminate up to half of entry-level white-collar jobs inside one to five years, with unemployment spiking to ten or twenty percent in the same window. The interesting feature of the claim is not its accuracy, which is unknowable, but its source: the head of a frontier lab said it in public, with his own name attached, while shipping the product that causes the outcome. CEOs do not loudly predict the obsolescence of their customers' cost structure unless they have already priced in the reaction. Amodei did it anyway, in language Axios picked up and every newsroom with a business model ran. The window in which a frontier lab can deny its own outputs has closed.

The 2025 and 2026 earnings calls tell the same story in the institutional dialect. Meta, which cut roughly five percent of its workforce in early 2025 under the "low performer" framing, has since replaced the language with explicit productivity-per-engineer metrics tied to AI tooling. Microsoft cut nine thousand positions in mid-2025 while posting record cloud margins, and the CFO, asked to explain the combination, cited capacity substitution in plain English. Marc Benioff at Salesforce declared publicly that the company would not hire software engineers in 2025 because Agentforce had made the existing engineers more productive. IBM's Arvind Krishna paused hiring for roles his team projected AI could perform. Google reorganized entire product lines around the presumption that engineering headcount per shipped feature would be halved.

None of this is hidden. The labor market is not quietly repricing. It is loudly repricing.

Why do all of these professions feel the same shock at the same moment?

Postmodality, Defined

There is an obscure subfield of the arts that had a name for this pattern before the pattern spread. It was called postmodality, and it referred to the condition that arises in a creative discipline when the tools used to produce the work stop defining the work. The term is awkward, which is probably why it did not travel. (I called this condition postmodality in a 2019 dissertation; the word has not caught on, but the condition has.)

The relevant definition comes from that document: "Postmodality questions the division between mediums and posits the long-term superiority of a portable aesthetic over technical knowledge" (dissertation, p. 12). The force of the claim, which was applied at the time to computer musicians confronting the collapse of their field into a broader soup of digital art, is independent of the art context. It is a claim about how value behaves when the tools get cheap.

A medium, as the term is used here, is the discipline-specific container in which work happens. Oil paint. The symphony. C++. The legal brief. The radiological image. The spreadsheet. Each medium has its own rules, its own ways of being wrong, its own canon of prior work, its own way of training a new practitioner. Mastering a medium is expensive. It takes years. For most of the modern era, the economy has been willing to pay for that mastery, because there was no other way to produce the work.

A portable aesthetic, by contrast, is an artistic sensibility that is abstracted from any particular medium. It is the taste, judgment, and sense of what-is-good that a practitioner carries into any medium they pick up, including mediums they did not train in. The original definition, in a footnote of the dissertation (p. 12, fn. 10), reads: "an artistic sensibility that is abstracted from the concerns of any one particular medium, thus is portable to any medium." That footnote has done more work in the seven years since its publication than the rest of the document combined, though only its author noticed.

The contrast, once it is named, is easy to see. High craft plus low taste is the dying center of every creative profession. It is the stock photographer who can operate a 5D Mark IV perfectly but cannot tell you what a good image is. It is the lawyer whose briefs are grammatically spotless but whose arguments are structurally inert. It is the engineer who writes textbook C++ for the wrong program. High taste plus low craft, until very recently, was called "not having made it yet." The artist who knew what was good but could not yet produce it, the analyst who saw the real question but could not build the model, the writer whose ideas outran his sentences. Now, because the tools will do the making, that same practitioner is called employable.

Richard Wagner wanted a total work that unified music, drama, and staging into a single expression, and he called the ambition Gesamtkunstwerk. He wanted, in his own phrase, the Fellowship of all the Artists (Wagner 1892, quoted dissertation p. 15). The tools to build such a thing did not exist in his lifetime. The ambition waited. GPT-class systems and their successors are, for the knowledge worker, the mass-market version of that ambition: a single system in which the individual practitioner composes across mediums they did not train in. Wagner's problem is now the median problem of a Monday morning.

The Euphemism Treadmill

Disciplines whose substance is dissolving tend to get renamed, and the renaming is typically the clearest public signal that the substance is gone. Steven Pinker called this the euphemism treadmill: the phenomenon in which new words attach themselves to fading categories, become tainted by association with the category they replaced, and get replaced in turn, without any of the substance actually changing (Pinker 2002, quoted dissertation p. 4). Garbage collection becomes sanitation becomes environmental services. Undertaker becomes mortician becomes funeral director. The procession is orderly, and its orderliness tells you nothing has been solved.

The arts have been on this treadmill for decades. "Computer music" became "electronic music" became "multimedia art" became "experimental media" became "creative technology," and will become something else again before this sentence is filed. Every rename has coincided with a further loss of the discipline's distinctiveness. No new category has captured the center, because the center is not a category anymore. It is a continuum. The dissertation was blunt about it: "They will cease to exist solidly as individual fields, instead integrating into a seamless continuum of practices" (dissertation, p. 4).

The same process is now visible in the labor categories that tech intellectuals have historically treated as load-bearing. "Software engineer" in 2019 meant a specific thing: a person who could hold a medium-complexity system in their head, produce working code in at least one language, and ship features on a team. In 2026 the category has split. On one end are the small number of practitioners who architect systems that AI tools then fill in. On the other are the larger number of practitioners who do what is now called vibe coding, a term Andrej Karpathy coined as half-joke and half-prediction: describing outcomes in natural language, letting the model produce the code, and accepting the result if it runs. Cursor, Claude Code, Devin, and the next three tools that will be named before this essay is published have all compressed the craft into a narrower band. The valuable worker is no longer the person who writes the most correct syntax. It is the person who can specify taste, architect intent, and translate across representation layers.

This is a portable aesthetic for code. The naming will not be this clean, because no naming is clean while it is happening, and because the community has a strong incentive to keep calling its most prestigious role "software engineer" even after the role's content has shifted. The euphemism treadmill will produce three or four candidate terms in the next five years. None of them will describe the substance well. The substance will remain what it is: an aesthetic discipline dressed in the old coat of an engineering one.

The pattern is not confined to technology. Investment banking's entry-level class, which for generations existed to assemble pitch decks, format financial models, and shepherd documents between senior bankers, is being quietly reconstructed into a supervisory layer above a set of AI workflows. The title remains analyst. The substance of the role is now closer to what senior associates did in 2015. A cohort whose entire professional identity was built on the grind of the first two years out of undergraduate is finding that the grind was the medium, and the medium was what they were being paid for, and the medium is now automated. The title will change, eventually, because the title can no longer describe the work. What it will change to is anyone's guess; the list of candidate replacements circulating on internal Slack channels already runs to eight items, none of them descriptively accurate, because description is not what renaming is for. Renaming is for postponing the reckoning.

A prior generation of technologists called this pattern in real time. David Zicarelli, the founder of Cycling '74 and the architect of the Max/MSP environment that a generation of computer musicians built their careers on, gave a keynote at the 2001 International Computer Music Conference in which he predicted that his own field's technical obsession "may well end soon, and with it, we will see the decline into increasing irrelevance of the institutions that have supported it" (Zicarelli 2001, quoted dissertation p. 6). Filed alongside Moore's Law, Gibson's cyberspace, and Lanier's haptic glove, the Zicarelli sentence belongs to a small corpus of claims made by insiders who saw their own discipline's coming dissolution before the general public had any reason to care.

The Case of Wagner, Gibson, and Lanier

There is a pattern that recurs in the archive of technological change: a practitioner in some marginal discipline articulates the shape of a condition before the material base exists to support it. The articulation does not produce the condition. It sits on a shelf. When the condition arrives, the articulation is retrieved, and the retrievers are surprised.

Richard Wagner in 1849 wrote The Art-Work of the Future, in which he demanded a total-medium artwork uniting poetry, music, staging, and audience into a single experience. His contemporaries considered the ambition grandiose and, for the practical purpose of producing such a work in 1849, correctly so. No theater of the time could execute what Wagner was asking for. Bayreuth took decades to build. The opera house is itself part of the argument: Wagner's ambition required its own infrastructure, because the infrastructure that would have sufficed did not yet exist. More than a century later, film and then digital games delivered, to a mass audience, the synthesis Wagner could only gesture at. His claim had been correct on substance and wrong on timing.

William Gibson published Burning Chrome in 1982, and in it used the word "cyberspace" to describe a shared hallucinatory network of information, experienced sensorily by its users, in which criminal acts were performed with real consequence. The internet of 1982 could barely move a text file. The claim that there would one day be a consensus-hallucination economy in which trillions of dollars of fraud, art, theft, and commerce were transacted inside a shared symbolic space was received as the conceit of a science fiction novelist. Gibson's vocabulary became the default vocabulary of the 1990s tech press inside of a decade.

Jaron Lanier, throughout the 1980s at VPL Research, designed the DataGlove and the EyePhone, and more importantly designed a way of thinking about virtual reality as a medium with its own grammar rather than a 3D extension of the monitor. VPL was a small, strange, influential company that went bankrupt on schedule for a small, strange, influential company. The devices it prototyped at roughly fifty thousand 1988 dollars are, with modification, now available as consumer peripherals for a few hundred. Lanier's claim was correct. The timing was wrong, and the timing was wrong by enough that the original company could not capitalize on its own accuracy.

What is instructive about all three figures is not their individual prescience, which in each case was modest on its own, but the institutional location in which the prescience was produced. Wagner was an opera composer, not a theorist of industrial capitalism. Gibson was a novelist, not a telecommunications engineer. Lanier was a musician who had wandered into a computing lab, not a product manager with a roadmap. None of them was incentivized, by position or profession, to tell the truth about where the adjacent fields were going. That freedom from incentive is what made them tell it. An engineer at AT&T Bell Labs in 1982 could not have written the opening of Burning Chrome without losing his job, and in any case would not have been believed if he had. The claim had to come from outside the incumbent institutions of its time. When the inside finally picked it up, it did so through acquisitions, reorganizations, and obituaries, rather than through any earlier admission that the outside had been right.

Each of these figures did the same thing. They articulated the structure of a condition before the tools to produce the condition existed at consumer scale. None of them was right about when it would arrive. All of them were right about what it would look like. Dissertations, novels, and failed company prototypes are where the forward-looking claims get stored, because the institutions with revenue models have no incentive to store them; only an incentive to ignore them until the shelf itself collapses.

Prophecy is cheap when it does not threaten the institutions that read it. That is why the most predictive claims are almost always made by outsiders, filed by insiders, and cited, posthumously, by the next generation's consultants.

What the Portable Aesthetic Actually Is

A diagnosis without a concept is a complaint. The portable aesthetic has four properties. The first is abstraction from tools. A practitioner with a portable aesthetic can describe their taste without naming the medium it operates in. A good writer is not a good sentence-assembler; they are a good argument-constructor who happens, at this moment, to be using sentences. A good designer is not a good Figma operator; they are a person with views about attention and hierarchy, who happens, at this moment, to be using Figma. The moment the practitioner cannot describe their own taste in medium-agnostic terms is the moment their value is tied to a medium that may not survive the next tool cycle.

The second is translation fluency. The practitioner can carry a principle across mediums without losing its force. A rhythm learned in music can be carried into film editing. A sense of negative space learned in graphic design can be carried into product copy. A structural intuition learned in one legal domain can be carried into another. This is the property that current models cannot yet trivially reproduce, because they do not have taste, only surface statistics about the distribution of taste in their training data. The window in which translation fluency is compensated above average is probably narrow, measured in years rather than decades, but while it is open it is wide and deep.

The third is judgment under ambiguity. When two correct outputs exist, the practitioner picks the right one for the context. Current models do not pick; they sample. Sampling is cheap. Picking is not. The picker knows, from some internalized model of the reader, the client, the audience, or the institution, which of the plausibly correct outputs is the one that will land in this room. The picker's judgment, in a world of cheap output, is the scarce resource. The economic shift of the next decade is the shift from paying for outputs to paying for picks.

The fourth is source-width. The practitioner reads widely, across fields, on purpose. This is the dissertation's own principle: "Sourcing widely is the most important design principle for computer musicians foraying into VR territory since this medium emerges out of a number of different mediums" (dissertation, p. 19). That sentence was written about a marginal arts discipline in 2019. It is now, without modification, the operating principle for any knowledge worker in a tool regime where the tools span mediums faster than any single practitioner can master them.

Consider the concrete case of a product designer in 2026. A decade ago the role required proficiency in Sketch or Figma, a taxonomy of pattern libraries, familiarity with React component hierarchies, and the ability to hand off specs that engineers could build without clarification. All of that is now automated, commoditized, or in the process of becoming so. What remains is the designer's ability to describe, in language the model can operationalize, what the interface should feel like; to carry a coherent rhythm of interaction across product surfaces that used to be built by different teams; to pick, among the seventeen visually valid variants the model returns, the one that will not insult the user; and to have read enough outside their own field, in film, in prose, in industrial design, to know which variant that is. That designer keeps their role. The one whose value proposition was the Figma file has already lost it. The same decomposition can be performed on any white-collar job whose title contains a tool name.

These properties are not new. They are the properties of every craftsperson who has survived a tool transition since movable type. What is new is their uneven distribution. A generation of workers was trained, explicitly and by selection, to specialize in tools, because specialization paid a premium and because the institutions that trained specialists were the institutions that collected tuition. That bet has been called. The premium is now on the portable aesthetic, and the workers who have it are a minority of the workforce, because the workforce was not trained to have it.

A note against the soft reading of this argument. The portable aesthetic is not a consolation prize for people who refuse to learn the tools. The aesthetic is hard-won, and is only visible against deep fluency in at least one medium. You do not acquire taste by refusing to practice a craft. The 2019 dissertation is explicit about this, in a passage describing the author's own training: only after learning the rules of European modernist composition was he able to depart from them in any serious way (dissertation, p. 13). The same is true for every practitioner in every field. Master one medium thoroughly enough that the pattern underneath becomes legible, and the pattern is your portable aesthetic. Refuse to master any medium, and the aesthetic is not yours; it is a story you tell yourself about why you were not hired.

Coda: The Shelf

Return to the epigraph. It was written in 2019, on a timeline that expected the audio tools to commoditize first and the general tools to follow, which is approximately what happened. The dissertation containing it was filed in the UC Irvine library, uploaded to eScholarship, assigned a permalink, and forgotten. Its author was hooded at graduation, took a photograph with his family, and went to work as a programmer, because the labor market was not, in 2019, paying for the portable aesthetic. It was paying for the medium.

This is the ordinary life of an accurate prediction. Prophecy that does not threaten incumbents is filed, not circulated. The institutions that train prophets do not have a business model for distributing prophecy. They have a business model for distributing credentials. The credentials are distributed on schedule. The prophecy sits. A generation later, when events confirm the prediction, the institution collects the reputation for having produced it, and the prophet, if he is lucky, is invited back to give a keynote. More often he is not alive to be invited. This is not a complaint about the institution. It is a description of how institutions work.

Seven years later, the labor market is repricing to match the thesis. The repricing is not being performed by the people who read the dissertation, because almost no one read the dissertation. It is being performed by firms that have, on their own, discovered that the premium on technical craft is compressible, that the premium on taste is not, and that the ratio between the two has flipped in a span of years rather than decades.

The mechanism is not complicated. Each firm discovers, independently, that the worker it is willing to pay double for has certain properties. The firms describe the properties in the vocabulary available to them, which is the vocabulary of the professions they are hiring into. A consulting firm calls the property judgment. A design shop calls it taste. A law firm calls it commercial sense. A venture fund calls it pattern matching. None of these firms knows, because none needs to know, that they are all pricing the same latent variable. The variable does not require a common name to be priced. It only requires enough independent discovery by enough firms for its market clearing rate to stabilize. That stabilization is now happening, in public, in the compensation bands of every profession the labor market has not already eliminated. Academic concepts get overtaken by this kind of discovery. The concept does not need to be read to be priced, and once it is priced, nobody goes looking for the reading.

There is a small number of practitioners, in every discipline, who recognized the turn early. They do not typically announce themselves. They do not need to. They are identifiable by what they built before it was obvious, and by how little they flinch now that the obvious has arrived. Some of them have dissertations on a library shelf. Some of them have notebooks. Some of them have nothing on paper at all; only a record of the work, visible only to those who also did the work.

The portable aesthetic is now the governing concept of the knowledge-work labor market, whether or not any market participant uses the phrase. Some readers of this essay will have recognized the concept from their own practice. Most will have recognized it from its absence in theirs. The gap between the recognition and the market's pricing of the recognition is the only arbitrage still open in this cycle, and it is closing.