In 1930, newspaper readers across the United States and Canada kept meeting the same villain: a robot. It fed musical instruments into a meat grinder and lunged for the helm of “musical culture”. The ads came from the Music Defense League, which the American Federation of Musicians (AFM) formed that year and which spent over $500,000 on the campaign. Its enemy was recorded sound: “Tho’ the Robot can make no music of himself,” one ad read, “he can and does arrest the efforts of those who can.”
The cause was concrete. Silent films needed live players, and by 1926 about 22,000 musicians worked in American theatre pits, a fifth of the AFM’s members. Sound film arrived in 1927, and within two years, in the union’s own later account, most of them were unemployed. The members paid for their campaign with a 2 percent tax on themselves. The pit orchestras did not come back, and recorded music became the ordinary way to hear music: in 2025 it was streamed on demand 5.1 trillion times.
That sequence has repeated in programming and in music. This essay sets the clashes out in order, with dates, and reads the pattern across them: what triggered each fight, what form it took, what decided it and what became of the old’s leverage. The pattern now applies to the masters of the digital audio workstation, who have become the old. Their first clash, with prompt-to-song apps, has begun. The larger one will probably come with general AI agents that make music through an instrument, because prompt-to-song runs into a limit no model removes: no language can say exactly what a sound is.
The priesthood
In 1954, almost all programming was done in machine language or assembly language: the processor’s own instructions, written out one at a time. IBM’s history of FORTRAN, written by the project’s leader, gives the economics that started the project: “the cost of programmers associated with a computer center was usually at least as great as the cost of the computer itself”. FORTRAN let scientists write formulas and had a compiler translate them into machine instructions.
The skeptics had evidence. Earlier “automatic programming” systems “slowed the machine down by a factor of five or ten”, and their promoters had claimed “almost human abilities to understand the language and needs of the user”. The FORTRAN team accepted the test: if its compiler produced programs “only half as fast” as hand-coded ones, “acceptance of our system would be in serious danger.” It treated the compiler, not the language, as “the real challenge”, and FORTRAN shipped in 1957.
The resistance was also about who could program. The same author described the programmers of the 1950s as “a priesthood guarding skills and mysteries far too complex for ordinary mortals”, who “wanted and got simple mechanical aids for the clerical drudgery which burdened them, but they regarded with hostility and derision more ambitious plans to make programming accessible to a larger population.”
Then FORTRAN became the old. In July 1983 the trade magazine Datamation printed “Real Programmers Don’t Use Pascal”, a parody in which real programmers write in FORTRAN or assembly and users of newer languages are “quiche eaters”. In 1997 a conference paper asked “Will C++ be faster than Fortran?” In 2007 an article presented Python to scientists as “an excellent ‘steering’ language for scientific codes written in other languages”, “often fast enough to be immediately useful”.
Python answered the speed charge by delegation. SciPy “wraps highly-optimized implementations written in low-level languages like C and C++”, its linear algebra is “always compiled with BLAS/LAPACK support”, and LAPACK “is written in Fortran 90”. In 2024 Python became the most used language on GitHub, “overtaking JavaScript after a 10-year run”. The old languages became the engine room of the new one, and Fortran added object-oriented programming in its 2003 standard. The latest round began with AI coding assistants: in Stack Overflow’s 2025 survey, 84% of developers were using or planning to use them, while more distrusted their accuracy (46%) than trusted it (33%).
Canned music, synthesizers and the sync button
I use “DAW” and “instrument” interchangeably, because the DAW is the latest in a lineage of sound generators: plucked strings such as the two-stringed dombra; electric and electronic instruments; synthesizers such as the TB-303, released in 1981 to replace the bass guitar, a commercial failure discontinued in 1984, whose cheap second-hand units became a foundation of acid house; techno, made with drum machines, sequencers, synthesizers and DAWs, whose hard techno now pushes the frontier; and the DAW itself, which holds all of these in one program. In music, the clashes between these generators were fought by unions, so they left dated records.
On August 1, 1942, the AFM stopped its members from making commercial records, in a dispute over royalty payments from the record companies. Decca settled in September 1943, “agreeing to make direct payments to a union-controlled ‘relief fund’”. The union refused a wartime presidential request to comply with a War Labor Board directive, and RCA Victor and Columbia gave in on November 11, 1944. The ban did not stop records; it attached a fee to them.
After the Moog synthesizer reached commercial music in the late 1960s, its ability to imitate strings and horns threatened session musicians’ jobs, and the AFM negotiated a ban on its use in commercial work in the United States. The instrument’s inventor thought the union imagined that “all the sounds that musicians could make somehow existed in the Moog”, so that pressing a button bearing a famous violinist’s name would bring out “the most fantastic violin player”. The ban ended when a synthesist persuaded the AFM that the instrument demanded skill, and “synthesizer player” became a membership category.
In 1980 the LM-1, a drum machine that played sampled drums, was released. It was feared that it “would put every session drummer in Los Angeles out of work”; many of the city’s top session drummers bought drum machines and learned to program them “in order to stay employed”. Labels increasingly use drum programming “to undercut the costly expense of studio drummers”.
In 1982 a pop tour of Britain used synthesizer players instead of an orchestra, and in May a meeting of the Musicians’ Union’s Central London Branch passed a motion for an outright ban on synthesisers. It never became union policy. In November the executive passed “a much more nuanced resolution”, synth players briefly formed a Union of Sound Synthesists, and the music paper NME called the branch “MU loonies”.
In 2003, in contract talks, Broadway’s producers proposed cutting the minimum orchestra from 24–26 players to as few as seven, with a “virtual orchestra” filling the gaps, and threatened to replace all musicians if they struck. AFM Local 802 struck from March 7 to 11; actors and stagehands honoured the picket lines. The settlement cut the minimums to 18–19 for ten years.
Among DJs, the “sync” button that matches tempos automatically made beatmatching by ear optional and turned using sync into an accusation. The British DJ group Girls Don’t Sync chose its name as a rejection and reclamation of the claim that women DJs reach for it.
The DAW’s own arrival left closures rather than a fight; I have found no union campaign against it. When New York’s Hit Factory announced its closure in 2005, the report cited “the rise of inexpensive digital audio workstation equipment”, which “has allowed high-quality recording to be accomplished in virtually any environment by non-professional recordists”.
The clash has started early
The masters of the DAW are now the old. The DAW carries what I call the classical studio idea, a song built as tracks on a timeline and sent through a console and a rack of effects, and mastering it takes years. Those years are leverage. Their first clash is with prompt-to-song apps, and it has run the familiar sequence in two years.
In April 2024 more than 200 artists signed an open letter warning that “unchecked, AI will set in motion a race to the bottom that will degrade the value of our work”. In June 2024 a sample platform surveyed 1,107 producers recruited through social media and reported that 25% used AI, mostly to split mixes into parts (73.9% of users) and for EQ and mastering (45.5%), while 3% used it to generate full songs. Of the non-users, 82.2% gave artistic and creative reasons (“I want my art to be my own”), 34.5% quality and 10.2% copyright. On June 24, 2024, the RIAA, for Universal, Sony and Warner, sued Suno and Udio for training on unlicensed recordings.
On February 25, 2025, against proposed changes to British copyright law for AI, 1,000 musicians released Is This What We Want?, recordings meant to represent “an absence of music and musicians”, whose track titles read “The British Government must not legalise music theft to benefit AI companies”. In October 2025 Universal settled with Udio; in November Warner settled with Udio and then with Suno. The settlements came with licences: new models trained on licensed music, downloads behind paid accounts.
The licences caused the next fight. On June 5, 2026, the AFM sued Universal and Warner, arguing that the deals triggered the “new use” provision of its collective bargaining agreement and licensed recordings its members worked on “without compensation or credit”. In October it launched “Human Made, Human Paid”, a campaign for consent, credit and compensation, ahead of contract talks with the major labels. That is the 1942–44 sequence in two years: resistance, lawsuits, a fee attached to the new thing, then a fight over who receives the fee.
The pattern
Side by side, the ten clashes read:
| Clash | What got cheaper | Resistance | What decided it | The old’s leverage afterwards |
|---|---|---|---|---|
| FORTRAN, 1954–57 | Programming, then as costly as the computer | Skepticism, derision | Compiled code close to hand-coded speed | The priesthood’s gate opened to scientists |
| C++ and Python, 1983–2024 | Writing scientific code | Ridicule, the speed charge | ”Fast enough”, with compiled cores | Old languages kept as the engine room |
| Sound film, 1927–31 | Film accompaniment | A $500,000 ad campaign | Theatres no longer needed pit players | Most pit jobs gone within two years |
| Recording ban, 1942–44 | Recorded performances | A two-year ban on recording | The union’s hold on recording musicians | Converted into payments to a union fund |
| Moog, late 1960s | Strings and horns from one player | A ban in commercial work | The instrument needed skill | A new membership category |
| LM-1, 1980 | Drum tracks | Fear; no organized action found | Drummers learned to program | Kept by those who learned |
| Synthesizers in Britain, 1982 | Orchestral parts on tour | A branch motion to ban | The executive softened it; synth players organized | The ban never became policy |
| Virtual orchestra, 2003 | Pit orchestras | A four-day strike | A contract; actors and stagehands joined | Minimums cut from 24–26 to 18–19 |
| DAW, by 2005 | Studio recording | None found | Cheaper equipment, smaller budgets | Large studios closed |
| Generative AI, 2024–26 | Finished tracks | Letters, lawsuits, a silent album, a union suit | Settlements and licences | Licence fees; the AFM suing for a share |
Four regularities run through the table.
The trigger is a falling cost that a paid skill occupied. Accompaniment, hand-coded arithmetic, a string section, a drum track: each was a step between an intention and its result, and someone was paid for it. That payment is leverage, the rent a skill earns while the road is long. The Selection Pressure states the law behind it: when one cost falls, the rent relocates to whichever cost has not yet fallen. Each clash is the old holders’ response to that relocation.
Resistance takes the form the old’s organization allows. Individuals argue: the new tool is a toy, slower, not real, cheating. Organizations that control a bottleneck use it: a recording ban, a membership rule, a strike, a lawsuit. Holders with neither, like the large studios facing the DAW, close.
In none of the ten did argument decide the outcome. Three things did: whether the new tool met the old’s own standard (FORTRAN’s speed, Python’s “fast enough”); whether the old’s skill could move into the new tool (the synthesizer player, the drummer who programmed); and contracts where the old held a bottleneck (the 1944 fund, the 2003 minimums).
In none of them was the old’s leverage restored in full. It was converted into a fee, a membership category, a contract minimum or a new skill held by those who crossed over. Then the new tool’s users became the next old: FORTRAN’s programmers were mocking Pascal’s users by 1983, and DJs turned “sync” into an accusation.
No language for sound
Prompt-to-song is not the DAW’s successor, and not because it generates music. Its interface is a sentence, and no language can say exactly what a sound is.
The acoustics standard defines timbre, the quality that makes a clarinet and a violin sound different at the same pitch and loudness, only by exclusion: “that attribute of auditory sensation which enables a listener to judge that two nonidentical sounds, similarly presented and having the same loudness and pitch, are dissimilar”. A classic book on hearing answered that “this is, of course, no definition at all”, and called timbre “an ill-defined wastebasket category”. A 2019 survey of how people describe sound begins: “Because humans lack a sensory vocabulary for auditory experiences”, we borrow words from other senses, such as “bright, warm, sweet”. When Greek and English speakers rated instrument tones with thirty adjectives, their answers reduced to three dimensions: luminance, texture and mass.
Musicians work around the gap with sound itself. Producers compare their mixes with commercial reference tracks, which in Sound On Sound’s advice give “your ears and brain a more objective benchmark”. People who cannot name a song hum it. The exact languages for sound are not spoken ones: notation fixes pitch and rhythm but says little about the sound; “a MIDI file is not an audio recording”; a synthesizer’s settings specify a sound exactly, and nobody talks in them.
So a prompt can only choose a region. “A warmer bass” moves along one coarse axis among a thousand basses, and the model fills in the rest with its own defaults. The prompt box is the button the AFM once imagined, and it returns the model’s guess at the label. Suno Studio, billed as a “generative audio workstation”, added a multitrack timeline and stems, and version 2.0 added MIDI editing and export and an AI chat. The leading prompt company built a workstation because the prompt was not enough, and its chat still routes intention through words.
The exact languages are where agents are strong. Music can already be written as code: TidalCycles is “a free/open source live coding environment for algorithmic patterns”, and its web-based relative Strudel runs in the browser. A synthesizer’s settings are a precise language of the same kind. An agent can write in languages that people do not speak, while the person judges by ear. Agents are also gaining ears. A 2025 paper introducing Music Flamingo notes that “music remains challenging due to its dynamic, layered, and information-dense nature”, and reports state-of-the-art results on more than ten music benchmarks.
When the successor arrives
My bet is that the general agents that mastered code will also be the producers. In October 2023, a public benchmark tested language models on 2,294 real GitHub issues from 12 Python projects; the best model resolved 1.96%. By December 2025 the leaderboard’s top agents resolved 52.6% of the full set and 79.2% of its Verified subset of 500 issues. Music shares a good part of what made that possible, structure, pattern and logic in an editable project, and adds a dimension code does not have: sound, and an ear to judge it. What separates music from code is that dimension, not a number of years. The general models that now write code will make music and synthesize sounds through an instrument, with sound generators written as code and neural sound models as their tools, and they will build their own tools to hear what works and what does not, as they now write tests for their own code. That, rather than prompt-to-song, is the new that the masters of the classical studio idea will face.
It changes the order of learning. The DAW asks you to learn first and build later; an instrument with an agent in it lets you ask, hear, adjust and continue, learning while you build. That loop is older than any studio. In the art of the dombra kuy, which UNESCO inscribed in 2014, “aspiring and talented musicians are apprenticed to masters from the moment a child demonstrates an interest”.
The pattern predicts the response: first the charges (a toy, a generic sound, not real production, cheating), then rules and contracts, then fees, then absorption, with many of today’s masters among the first to learn the new instrument, as Los Angeles session drummers learned to program the drum machine. It also offers a test. In 2024, producers’ main stated reason for refusing AI was artistic and creative control. An agent working inside the instrument, where every change can be heard, kept or undone, removes that reason. If resistance fades then, the objection was control; if it continues, it was leverage. What such an instrument needs is the subject of De DAW of the Future.
What could prove this wrong
- Language may be enough. If people reach the sound they mean by conversation as quickly as by hearing and adjusting, the argument from language fails; time and attempts can measure it.
- Music may not follow programming. Code has tests; music has only the ear. If agents’ musical ability does not climb as their coding ability did between 2023 and 2025, the bet fails.
- The fight may be about consent and pay. If resistance falls as licensed, consented models spread, leverage was the smaller cause.
- Rules can hold. Broadway’s minimums survived 2003, smaller but intact. Contracts and laws can slow the new.
The robot in the 1930 ads “can make no music of himself”, and recorded sound made none; people made it, and many more of them could. Each clash was between a want and a toll. The want took the shorter road, the toll-keepers fought, and then they learned the road and kept the next toll.