Every musician knows the moment. After weeks, months, sometimes years of writing, recording, editing, overdubbing, mixing, second-guessing, remixing, mastering, unmastering, remastering, and listening to the same thirty-two bars until you are no longer sure whether you even like music, somebody finally says the magic words: This is the master. It is done. Nobody touch anything⊠(Insert sound of a âwhoopâ in the background)
For most of the history of recorded music, that moment has been sacred. The master was the final authority, the definitive version, the thing that went onto the record, tape, CD, download, streaming service, radio station, jukebox, movie soundtrack, or whatever technological container was currently separating musicians from their audience.
You made a record, and people listened to the record you made. They might love it, hate it, skip it, obsess over it, play it at the wrong speed, or destroy the EQ with their car stereo, but fundamentally everybody began with the same thing.
I am starting to wonder whether that entire idea is about to become optional.
Not because musicians have suddenly lost interest in finishing their work. We have plenty of other ways to avoid finishing things. The bigger change is that technology is beginning to make it possible for a piece of recorded music to remain variable after it has been released.
The recording can change.
The arrangement can change.
The duration can change.
The mix can change.
The instrumentation can change.
The way one section moves into another can change.
Eventually, the recording may respond to the person listening to it, the place where it is being heard, the time of day, the listenerâs activity, or any number of other inputs.
We are moving toward recorded music that may not play exactly the same way twice. That sounds like a technical novelty until you realize what it actually attacks: one of the central assumptions of the record business.
What happens when there is no longer one definitive recording?
Maybe the Recording Was Never Supposed to Be Frozen
We tend to treat the fixed recording as if it were some kind of natural artistic law. The musician creates the music, captures the performance, finishes the mix, and there it is forever.
But that is not how music behaved for most of human history. Music happened, disappeared, and happened again. A musician played a piece one night and played it differently the next night. Tempo shifted. Phrasing changed. Solos changed. Musicians responded to the room, the audience, each other, and whatever kind of day they were having.
Then recording technology arrived and performed a miracle. It allowed us to capture a musical event and repeat it exactly.
That was extraordinary, but somewhere along the way we confused the capability with the definition. Because a record could repeat exactly, we gradually came to believe that recorded music was supposed to repeat exactly.
Vinyl reinforced the idea because grooves do not improvise. Tape reinforced it because the same magnetic information passed the heads every time. CDs replaced analog information with digital information but preserved essentially the same deal. Streaming removed the physical object entirely and somehow managed to reproduce the exact same philosophy: here is the file, press play.
Different container. Same assumption.
Brian Eno saw the limitation a long time ago. His experiments with generative music were based on a fundamentally different idea: instead of composing every final musical event, you could compose a system capable of producing musical events.
That distinction is enormous. In traditional recorded music, the artist decides what happens at 3:14. In generative music, the artist may instead decide what could happen at 3:14.
The work is finished, but the realization is not fixed.
That sounds contradictory only because we have spent a century thinking that finishing a piece of recorded music means freezing it.
Maybe it does not.
Brian Eno Has Been Warning Us for Decades
Enoâs generative experiments make much more sense now than they did when the technological machinery around them felt exotic. With works such as Generative Music 1 and later Reflection, he increasingly separated the idea of the composition from one particular playback of it.
The conventional version of Reflection exists as a finite recording. You can press play, hear it, and hear essentially the same recording tomorrow. But the generative version behaves differently. It can continue, alter itself, and produce different musical experiences within parameters established by the composer.
That gives us one of the most important distinctions in this entire conversation. A composition can have a fixed identity without having a fixed performance.
Jazz musicians already understand this. Classical musicians understand it. Anyone who has played the same song onstage two hundred times understands it. The piece remains the piece even though every performance is not identical.
Recorded music has generally been the exception.
Generative technology begins collapsing that distinction. The recorded work starts behaving more like the living music that existed before recording technology froze it.
The irony is beautiful. We spent more than a century perfecting technology capable of making music permanent, and now we are using vastly more powerful technology to make it move again.
What If the Song Is a Place?
The Bronze project explored this territory from another direction. Instead of treating a song as one linear path from beginning to end, the technology allowed musical elements to behave dynamically during playback.
Gwilym Gold used the system with Tender Metal. Jai Paulâs âJasmineâ also existed in a Bronze realization that could change from playback to playback.
The important word there is not ârandom.â
Randomness is easy. Put fifty audio clips in a folder and hit shuffle. Congratulations, you have invented a terrible afternoon.
The interesting problem is creating variation without destroying identity. If a song can take different forms every time you hear it and still unmistakably remain that song, then the artist has created something more sophisticated than a collection of interchangeable parts.
The song starts looking less like a road and more like a territory. There are boundaries. There are landmarks. Certain things belong there and certain things do not. The artist establishes the world, while the playback system finds different routes through it.
That idea could become incredibly important.
A conventional recording says, follow me.
A generative recording might say, enter here.
Those are very different invitations.
Massive Attack Made the Listener Part of the Mix
Massive Attack pushed the idea even further with the Fantom app. Their music could respond to information surrounding the listener, including movement, location, time, and biometric information such as heart rate.
Now we have crossed another line. The recording is not merely changing. You are helping determine how it changes.
Not because you suddenly became a member of Massive Attack, which I am sure comes as disappointing news to several people reading this. The artist still created the material and established the possibilities, but information about the listener could influence which version of those possibilities appeared.
That is personalization at a deeper level than a recommendation algorithm. Spotify recommending a different song because I listened to Peter Gabriel yesterday is one thing. The Peter Gabriel song itself changing because of how I am listening to it is something else entirely.
Imagine where this can go.
You are walking quickly through Manhattan and the rhythmic density increases. You listen late at night and the arrangement becomes more spacious. You are exercising and a section extends because your movement suggests you have not finished yet. You are lying in bed and the music gradually removes elements rather than arriving at the same hard ending every time.
The obvious reaction is that this sounds incredibly cool.
The second reaction should probably be:Â Do I actually want this?
We will get there.
When the Weather Joins the Band
Björk has experimented with a related idea in a more environmental direction. Her Kórsafn installation used choral material from across years of her work in a system that responded to conditions in the surrounding world.
Suddenly weather is not merely something happening outside while music plays. Weather becomes part of the input.
That represents another major conceptual shift. Traditionally, context changes our experience of music. A song sounds different in the car at midnight than it does in a restaurant at noon, even though the recording itself has not changed.
Dynamic music allows context to change the music itself.
Those two ideas sound similar, but they are not.
For the entire history of fixed recording, the listener adapted to the record. We changed rooms, speakers, moods, relationships, cities, and decades while the recording stubbornly remained itself.
Now the record can adapt to us.
That is seductive. It is also slightly creepy.
Music has survived perfectly well without checking my pulse before deciding whether the bridge needs another eight bars.
Endel and the Personalized Composition
Endel moves even further into adaptive sound. Its systems can respond to variables such as movement, time, environment, and physiological information, creating music that shifts according to the conditions surrounding the listener.
Collaborations with established artists make this especially interesting because the raw material begins with identifiable artistic choices. Miguel can create material, but the system determines how that material unfolds for different people under different circumstances.
That suggests a useful new division of labor. The artist does not necessarily compose every playback. The artist composes the field of possible playbacks.
That is not the same as handing authorship to the listener. I know we are living in the age when everybody with a smartphone apparently needs to be told that they are now a creator, filmmaker, broadcaster, curator, entrepreneur, thought leader, and possibly an astronaut, but listening to adaptive music does not suddenly make the audience the composer.
The artist still defines the musical language. The listener simply becomes another variable inside the system.
That distinction matters because we are going to hear a lot of nonsense about âco-creationâ as this technology develops. Moving faster on a treadmill and causing the percussion to become slightly more active does not put your name on the publishing.
The artist created the possibilities.
You triggered one of them.
Everybody can relax.
Stop Calling Everything a Song
Aimi went directly at the language problem. Its systems could take artist-created beats, loops, textures, and other musical elements and continuously assemble them into evolving streams rather than finite conventional tracks.
The company used the word experience instead of song.
Marketing departments have certainly committed worse crimes against language, but in this case the distinction actually makes sense.
What exactly is a song that can continue for three hours, rearrange itself, alter its structure, and produce a slightly different realization tomorrow?
Is it a track?
Is it a mix?
Is it a performance?
Is it software?
Is it a composition?
Is it an environment?
The record business is built around nouns developed for relatively stable objects. Song. Recording. Album. Master. Track. Mix.
The technology is beginning to produce things that behave more like verbs.
That is usually the moment when industries get confused, because industries love nouns. Nouns can be counted, packaged, copyrighted, licensed, delivered, monetized, reported, and placed into spreadsheets.
A constantly changing musical system is annoying.
Someone in accounting is going to need a meeting.
Video Games Have Already Been Living in This Future
Games have been dealing with adaptive music for years because games have an obvious problem that albums do not. Nobody knows exactly what the player is going to do.
A film score can be timed to the frame because the film is fixed. At 43 minutes and twelve seconds, the door always opens.
In a game, the player might open the door now, ten minutes from now, or spend forty-five minutes wandering around collecting something entirely pointless because apparently that is entertainment.
The music has to adapt.
No Manâs Sky provided an especially powerful example. 65daysofstatic created musical material designed to be broken apart and reassembled dynamically within the game. Instead of one score progressing in one predetermined sequence, musical components could be recombined according to what was happening inside an enormous procedurally generated world.
Then the artists did something wonderfully contradictory. Years later, they took material that had existed as fluid, generative music and shaped it back into fixed tracks for No Manâs Sky: Journeys.
That reversal matters.
Infinite possibility was available.
They still chose to make records.
Maybe that tells us something.
Maybe One Version Is the Point
This is where the future-of-music evangelists usually lose me. Every new technology arrives carrying the same sales pitch: more choice, more personalization, more versions, more interactivity, more control.
We rarely stop to ask whether more is actually the artistic objective.
Think about the recordings you love.
You know the drum fill.
You know the breath before the lyric.
You know the exact shape of the guitar solo.
You know when the strings enter.
You know the strange little sound in the left channel that probably happened accidentally and was smartly left alone.
You know what happens next.
That knowledge is not a flaw in the listening experience. It is part of why the listening experience becomes powerful.
We form relationships with recordings partly through repetition. The same musical event keeps happening while wechange.
You heard the record at nineteen. You heard it again at thirty-five. You hear it at sixty. The record stayed the same. You did not.
That is profound.
I am not convinced an adaptive mix that thoughtfully notices I am older and turns down the hi-hat would improve the experience.
Sometimes music is valuable precisely because it refuses to adapt to us.
Leave Kind of Blue Alone
There are musical performances that became important because one particular thing happened once and somebody had the good sense to record it.
We do not need a personalized version of every masterpiece.
I do not want Miles Davis generating a slightly different solo because my Apple Watch thinks I slept badly. I do not need A Day in the Life deciding that todayâs listener would prefer a more uplifting final chord. I do not want an algorithm shortening the second movement of a symphony because my historical completion rate suggests I lose interest around minute nine.
Some art has authority because the artist made decisions and then made us deal with them.
That relationship is increasingly unfashionable.
Everything is supposed to accommodate us now. Our feed accommodates us. Our advertising accommodates us. Our shopping accommodates us. Our entertainment is recommended according to our behavior. Our phones reorganize themselves around what we supposedly want before we even know that we want it.
Maybe music does not need to join the servants.
Maybe one of the most valuable things art can do is say, No. This is the piece. Meet me here.
Streaming Is Already Softening the Edges
The fixed recording has not disappeared from mainstream streaming, but the edges around it are starting to loosen.
Apple Music Sing allows listeners to alter vocal prominence on millions of recordings. The track remains the track, but one of the most important relationships inside the mix becomes adjustable.
AutoMix and similar transition technologies go further in another direction. The beginning and ending of songs are no longer necessarily inviolate because software can analyze tempo, key, rhythm, and structure, then create transitions between recordings.
Spotify is moving toward an even more consequential idea: licensed generative remix and cover tools built around commercially released music. Deals announced with major rights holders, independent music organizations, and publishers suggest that this is not merely a laboratory experiment.
That should get musiciansâ attention.
The streaming service used to deliver the record.
Now the streaming service is beginning to imagine delivering the ability to transform the record.
That is a major change in power.
The Master Starts to Look Very Different
For decades, record production eventually converged on one terrifyingly important file.
THE MASTER.
That file represented the end of the process. Everything else was a path toward it.
But imagine a future release package containing something entirely different: master vocals, instrumental stems, alternate performances, arrangement modules, tempo boundaries, key restrictions, spatial information, generative instructions, personalization parameters, approved remix behaviors, and perhaps an AI model capable of assembling those pieces according to rules created by the artist.
What is the master now?
Maybe the WAV file is merely one example.
Maybe the actual master is the system that knows what the piece is allowed to become.
That is the part of this idea I find genuinely exciting.
A musician could finish a work by defining its boundaries rather than defining every event.
The artist might decide that harmony never changes but orchestration can. The vocal performance is fixed but instrumental density responds to the listener. Duration can change but melodic material cannot. Sections can expand but never disappear.
Those are artistic decisions.
They are also compositional decisions.
Writing the rules of variation may eventually become as important as writing the notes.
Music Could Have an API
Stay with me here because this is where things get geeky and potentially fascinating.
Software developers build systems with rules about what other systems are allowed to request and what the software will return. Music may eventually acquire something conceptually similar.
A piece could contain permissions.
The bass line may be replaceable.
The lead vocal may not.
The instrumental section may extend between sixteen and sixty-four bars.
Tempo may shift within five percent.
The ending may have three approved forms.
Harmony may remain fixed.
Spatial placement may adapt to the listening environment.
The artist is no longer simply delivering audio. The artist is defining behavior.
That is very close to instrument design.
A violin maker does not determine the notes a violin will play. The maker constructs a system within which musical possibilities exist.
Generative recording could push composers and producers toward the same kind of thinking.
We would no longer simply ask, âWhat should happen here?â
We might ask, âWhat am I willing to allow to happen here?â
Those are not the same question.
Then the Business Side Gets Ridiculous
Naturally, the music industry will eventually have to count all of this.
That should be entertaining.
Existing recording identifiers assume that recordings are identifiable things. Create a materially different mix or edit and, under many circumstances, you have created another recording requiring its own identity in the commercial system.
Fine.
Now imagine a piece capable of generating one million legitimate variations.
Does each one get an identifier?
Of course not.
Does the system get the identifier?
Does the generated session get logged?
Does the underlying composition carry the identity?
Who gets paid if your personalized version contains twenty-three seconds more of the guitar part and thirty seconds less of the vocal?
If a generative system creates an eight-minute version for me and a four-minute version for you, did we consume the same recording?
Our current music-business infrastructure was designed to count objects.
Generative music creates possibilities.
Good luck with the spreadsheet.
Which Version Gets Reviewed?
The cultural questions may be even stranger.
Suppose I release an album that changes according to the listener.
Your version contains a longer piano section.
My version contains more electronics.
Another personâs version lasts forty minutes.
Someone else gets fourteen.
A reviewer listens and hates the percussion-heavy version. I listen and never hear that version at all.
What exactly did the reviewer review?
Which version gets submitted for an award?
Which version gets archived?
Which version goes into a box set twenty years later?
Which version gets played at the funeral?
That last question sounds melodramatic until you realize how much of our emotional relationship with recorded music depends upon its stability.
When someone says, âThis was our song,â they are talking about a repeatable object.
They remember that intro.
That voice.
That ending.
The song became a marker in time precisely because the marker stayed still.
What happens when the marker keeps moving?
Personalization Can Destroy Shared Culture
This may be the part of the whole idea that worries me most.
We already live inside individualized information systems. My social feed is not your social feed. My news environment is not your news environment. My recommendations are not your recommendations. Increasingly, we inhabit parallel cultural realities assembled around our individual behavior.
Music has resisted some of that because, while the playlist may differ, the recordings themselves remain shared.
Millions of people heard the same Beatles record.
The same Prince record.
The same Joni Mitchell record.
The same Stevie Wonder record.
That common object matters.
People can argue over the music because they are arguing about the same thing.
Hyper-personalized recordings begin eroding that.
My version.
Your version.
The workout version.
The sleep version.
The six-minute version.
The emotionally reassuring version.
The version optimized for the behavior some system has decided represents me.
At some point personalization stops being a feature and starts becoming cultural isolation.
We should probably notice before we enthusiastically optimize every piece of art until nobody is having the same experience anymore.
I Donât Need Music to Know Me That Well
There is something else underneath this that bothers me.
Technology companies are obsessed with eliminating friction.
They want software to know what we need before we ask.
They want entertainment to arrive already calibrated to our preferences.
That sounds wonderful if you are ordering paper towels.
Art is different.
I do not always want music that matches me.
Sometimes I want music that changes me.
I want to encounter somebody elseâs choices.
I want to hear a tempo I would not have selected, a sound I would not have chosen, a structure I find confusing, or a section I initially think is too long.
Great records have taught generations of listeners how to hear them.
That requires friction.
If music constantly reshapes itself according to our existing behavior, we risk building an enormous technological system dedicated to making sure nobody ever encounters anything they would not have chosen for themselves.
That sounds less like artistic progress and more like a very expensive comfort blanket.
The Artistâs Authority Is Up for Negotiation
This brings us to the part musicians should care about most.
Who decides what can change?
For most of recording history, the answer was ultimately the artist and producer. You could remix the record privately, change the EQ, cover it badly at a wedding, or sing over the instrumental, but the commercially released recording remained the artistâs statement.
Generative playback changes that contract.
If the fan can alter the recording, the artist needs to define the boundaries.
Maybe the listener can manipulate instrumentation but not the vocal.
Maybe duration can change but harmony cannot.
Maybe a solo can regenerate but a lyric cannot.
Maybe the music can respond to time of day but not biometric information.
Maybe absolutely nothing changes because the artist says this piece is finished.
That last choice needs to remain valid.
Technology has a bad habit of taking an available capability and quietly turning it into an expectation.
Once one artist offers ten customizable versions, somebody will inevitably ask why everybody else is so backward.
No.
Sometimes refusing a technological possibility is an artistic decision.
We May Release Both the Record and the System
There is another possibility, and I think it may ultimately be the most sensible one.
Maybe the definitive recording survives.
We simply add something beside it.
An artist could release the record: the fixed, final, intentional version that represents the statement.
Then the artist could release the system: a generative realization that expands, rearranges, adapts, or evolves within carefully designed boundaries.
One is the photograph.
The other is the landscape.
One captures the moment.
The other allows you to keep walking through it.
Brian Eno essentially arrived at this solution years ago. The fixed recording and the generative version do not have to compete because they serve different purposes.
That may be where all of this lands.
The mistake is assuming every recording needs to become fluid.
It does not.
The opportunity is realizing that some music may always have wanted to be fluid but technology forced it into a box.
Now we can finally open the box.
What Should Never Change?
That is the question I keep coming back to.
Not âWhat can AI change?â That list is going to become boringly long.
The more interesting question is what should never change?
What is essential to the identity of the piece?
The vocal?
The melody?
The harmony?
The form?
The exact performance?
The production?
The mistakes?
The silence before the final chord?
Every artist will answer differently.
That is exactly the point.
The future of recorded music may not be one enormous transition from fixed recordings to endlessly generated ones. It may be a new continuum between permanence and possibility.
Some works will be carved in stone.
Others will behave like weather.
Some will have one definitive performance.
Others may have ten thousand.
The artistic decision will be determining where the work belongs.
That is much more interesting than simply giving listeners another slider.
One Perfect Decision Still Matters
There is a strange arrogance in the assumption that infinite alternatives must somehow be superior to one great choice.
They are not.
Every serious artist knows that creation is largely an act of elimination.
You start with possibilities.
Then you reject them.
You change the chord.
Remove the verse.
Mute the guitar.
Shorten the introduction.
Rewrite the lyric.
Choose one vocal.
Commit to one tempo.
At some point, all those discarded possibilities collapse into a single statement.
That is what finishing something means.
AI can reopen the discarded doors.
Sometimes that will be extraordinary.
Sometimes it will simply demonstrate why we closed them.
We should not confuse the number of available possibilities with the quality of the result.
Ten thousand versions of a mediocre musical idea remain an impressive amount of mediocrity.
The Recording May Become an Instrument
This is where the idea finally connects to something larger.
If a recording contains rules, possibilities, alternate states, responsive behavior, and musical boundaries, then perhaps it is no longer merely a recording.
It has started behaving like an instrument.
You do not âplay backâ a piano.
You interact with it.
Every performance comes from the same underlying system, but the result changes.
Generative recorded music may begin operating similarly.
The artist designs the object.
The technology activates it.
The listener encounters one realization.
Tomorrow there may be another.
That does not destroy composition.
It expands what composition can mean.
The challenge will be resisting the industryâs inevitable desire to turn that possibility into another stupid feature nobody asked for.
Because the technology is not the art.
The artist still has to decide what the system is for.
We Spent a Century Making Music Permanent
Recording changed human culture because it defeated time.
A performance no longer had to disappear when the musicians stopped playing.
The dead could sing.
The young could hear people born before their grandparents.
A three-minute performance could become part of millions of lives for generations.
That permanence is one of the greatest artistic achievements technology has ever given us.
We should not casually throw it away because the new software can make the hi-hat different every Tuesday.
But neither should we assume permanence is the only valid form recorded music can take.
We now have another possibility.
A work can be authored without being frozen.
It can have identity without having one realization.
It can be finished without being final.
That is a genuinely new creative space for mainstream recorded music, even if experimental artists have been standing in it for decades shouting at the rest of us.
AI is not inventing the idea.
AI is making it difficult to ignore.
The Master Is Not Dead
I donât think the master recording is going away.
In fact, I suspect we are going to discover how much we value it precisely because alternatives now exist.
There will always be music where the exact performance matters.
There will always be records where changing anything makes them worse.
There will always be moments when the artist needs to say:Â No. This is the version.
Good.
I want that authority to survive.
But beside those fixed recordings, another kind of work is emerging. Music that breathes, expands, rearranges itself, responds, mutates, and produces different valid realizations without surrendering the identity established by the artist.
The important question is no longer whether technology can do this.
It can.
The real question is whether the music wants it.
We spent the twentieth century perfecting our ability to turn musical moments into permanent objects. We may spend part of the twenty-first learning when to let those objects move again.
That does not mean every record needs to become infinite, personalized, interactive, adaptive, biometric, generative, remixed, reconstructed, and optimized for my current blood oxygen level. Sometimes I want to press play and hear exactly what the artist decided I should hear.
But sometimes a piece of music may be bigger than one recording.
Sometimes the master may only be one photograph of a much larger world.
And when that happens, maybe the artistâs job is no longer to give us the final version.
Maybe the artistâs job is to build the world, define its laws, decide what can change, protect what cannot, and then open the door.
That is when the master stops being a file.
It becomes a systemâŠ