In the 1970s and 1980s, the great recording studios were among the most expensive rooms in the world. A console the size of a car, a wall of tape machines, an echo chamber, a staff of engineers and assistants, session players on call: if you wanted a record that sounded like a record, you rented the room. Careers were made in those rooms, and the Hit Factory in New York was among the most famous of them. And then, within a generation, the business was in retreat, and nobody attacked it. The Hit Factory closed in 2005, and its building became condominiums sold with the slogan “Live in the House That Rock Built.” EMI closed Olympic Studios in London in 2009 because it no longer made money, and in 2010 it put even Abbey Road up for sale, until the British government listed the building to stop developers turning it into flats. In 2020, the Grammy for Album of the Year went to a record produced in a small bedroom studio in Los Angeles, in an off-the-shelf DAW.
The mistake is to think the studios sold rooms, or tape, or consoles. They sold the finished record: the best way anyone had found to turn a musical idea into music other people could hear the way it was meant. When software served the same want far more cheaply, beginning with Pro Tools in 1991, the want quietly took the shorter road. Other forces pushed too: Manhattan rents, the collapse of record sales after file-sharing, labels in debt. But the decisive fact outlived all of them. A cheaper path to the finished record had appeared, and the consoles did not lose a fight. They were bypassed, and then redrawn, knob for knob, inside the software that replaced them.
That last detail is where this essay begins. The want is the constant and the means keep dying, as I argued in The Selection Pressure. The DAW did not abolish the studio; it shrank the studio to fit on a laptop and kept its shape. Now it faces the pressure it once applied.
Imagine hearing a song and wishing the bass would wait a little longer before it returns. The intention takes a second to form. Realizing it can take an evening: find the source, separate the parts, learn an editor, rebuild the passage, render it, compare it with the original, and hope you can find the file next week. The DAW is the most capable musical instrument ever built, and it still charges that evening as the price of the second. The reason is its shape. A studio is organized around jobs. An instrument is organized around the person holding it: a gesture goes in, a sound comes out, and the ear decides. The DAW inherited the jobs, and with them the rule that you learn the operation before you are allowed the intention. So the contest that matters is not between applications. It is over how far a person must travel between a musical intention and its realization, and what carries them across.
My thesis is simple. The DAW of the future will not be a better studio. It will be an instrument for shaping what you hear: a person and an agent working on the same editable music through the same commands, while the instrument supplies the senses, the work keeps its memory, and the ear keeps the verdict. Four distinctions carry the argument: studio and instrument, operation and intention, recording and project, evidence and verdict.
The pressure for the change is already visible in the numbers. In 2025, an average of 106,000 new tracks were delivered to streaming services every day, up from 99,000 in 2024. By September 2026, Luminate’s chief executive put the figure at close to 150,000. On Deezer, fully AI-generated tracks passed half of all daily uploads in June 2026, about 90,000 a day, while drawing only 1–3% of the platform’s streams. Of the 253 million tracks on streaming services at the end of 2025, 88% were played fewer than a thousand times that year.
Making a finished-sounding track has never been cheaper. Making music that matters, first to the person who made it and then to someone else, has not become cheap at all. The next instrument will be built in that gap. To see its shape, start with the person, not the software.
What people want from music does not change
Someone plays a familiar song after a difficult day. Someone else chooses it for a friend’s wedding. A third person hears a phrase and begins answering it on an instrument. The recording may be identical; the intentions differ. People use music to feel and to change how they feel, to express what they cannot easily say, to remember, to belong, to exercise taste, and sometimes to build a skill over years. Human wants change slowest. Objectives are more stable than actions, actions are more stable than interfaces, and implementations change fastest.
From those wants come three needs an instrument can serve: music that fits me now; music in which I recognize my own choices; and music through which another person answers me. Only the third needs another person, and only when someone actually wants to respond.
Two cautions keep this honest. Listening is a complete destination, not a failed attempt to create. And effort is not always a cost: people avoid the effort that stands between them and the music, and seek the effort that is the music, such as practising a rhythm or finding a sound by ear. A good instrument removes the first kind and protects the second.
That is the measure every tool in this essay is held to. Technology does not change what people want; it changes which wants have become affordable and which costs are left standing. For a generation, the tool that decided that for recorded music has been the DAW.
The DAW is a studio you carry
Every DAW on the market is, at heart, a recording studio drawn as software. The transport is a tape machine. The tracks are tape. The mixer is a console, the plug-ins are a rack of outboard gear, and the piano roll is a player-piano roll turned on its side. Each panel stands for a job that a person once did.
That inheritance was a triumph. Recording made a performance repeatable. Multitrack made its parts revisable. Software put the whole construction on a personal computer. But what software collapsed was the cost of equipment, not the cost of knowing. Affording a tool and realizing an intention are different kinds of access. The second kind is the distance between operation and intention, and a studio cannot close it, because a studio is made of operations.
The history of music technology is a sequence of such partial victories. Notation let musical instructions travel without the performer; literacy, teaching and publishers remained the gates. Recording let people hear music without its players in the room; financing, manufacture and promotion became the gates. MIDI, agreed in 1983, let performance data travel between machines from different makers, but it carried the instructions, not the music. Each wave put a new power in people’s hands and left a gate standing.
Tools also acquire meanings their makers never planned. Roland released the TB-303 in 1981 as a bass accompaniment for guitarists. It failed at that job and was discontinued in 1984. Cheap second-hand units then became the squelching voice of Chicago acid house, beginning with a record called Acid Tracks. Nobody can design a scene. The most an instrument can do is leave room for one.
A chain bought in layers
Look at what an electronic producer actually pays for, and the DAW turns out to be one link in a longer chain:
- the engine: the synthesizers that make sound;
- the content: presets, samples, loops and patterns, per genre;
- access: libraries that supply and search sounds on demand;
- know-how: lessons, and other people’s project files;
- assembly: turning parts into a track, alone or with others;
- release: labels, streaming services and direct sales;
- performance: the DJs, clubs and festivals where music meets a room.
Each layer has its own seller and is someone’s livelihood: instrument makers, sound designers, pack makers, tutors, engineers. The producer pays twice for every layer: once in money, and once in learning.
The DAW sits at the assembly layer, the workbench where parts bought elsewhere are put together. That is why it has been so stable for so long: it is the centre of a chain whose other links it does not own.
Integration alone is not the answer, and neither is price. Free, integrated workstations already exist at scale: BandLab, a free studio for the browser and the phone with loops and collaboration built in, passed 100 million users in 2024. What would be new is an instrument that joins the layers and acts on the work itself. Where the cost now sits decides whether that is worth building.
When sound is abundant, scarcity moves
Each wave above left a gate standing, and the rent moved to it: from players to distribution, from equipment to learning, from delivery to attention. That is the law of The Selection Pressure: when one cost falls, the rent relocates to whichever cost has not yet fallen. Music is running the experiment at full speed.
Generative models have made starting material abundant; the upload figures above are what abundance looks like, and they carry a warning. Cheaper production does not make sources cheaper to obtain. Easier delivery does not bring an audience. More output does not show that any of it matters to anyone. Deezer estimates that up to 85% of the streams that fully AI-generated tracks received in 2025 were fraudulent.
If sound is no longer scarce, what is? For a person trying to realize a musical intention, four things remain expensive:
- Understanding and control: reaching the part you mean, knowing what a choice will do, and hearing whether it helped.
- Learning and attention: gaining a capability without spending more on setup and evaluation than the outcome is worth.
- Reliable material: finding, preparing and keeping the sounds the work depends on.
- Continuity and trust: returning to the work, tracing where each piece came from, and handing it on in a state someone else can continue.
These are not features. They are where the cost now sits, and the next instrument will be judged by how much of that cost it removes. The most visible answer to abundance removes almost none of it.
The prompt is not an instrument
That answer is the song generator: type a description, receive a finished track. It is a remarkable technology, and the record industry has decided to live with it. Universal Music settled its lawsuit against Udio in October 2025 and agreed to build a licensed service with it. Warner Music settled with Suno a month later and announced a partnership. Prompt-to-song is being absorbed into the old chain as a catalogue business.
That settles a legal question. It does not settle a musical one.
A prompt steers only approximately. Words fall short of what a person hears: “a warmer bass” describes a thousand basses, and people usually find the one they meant by hearing it. The deeper problem is exactness. A generated recording is accepted or rejected whole. You cannot keep the melody you love, change the rhythm under it, and then nudge a single hi-hat, without asking the machine to roll the dice again and hoping it remembers what you liked.
Generation becomes part of production only when it has an editable target to generate into: a project that the person and the agent can both open, hear and change, so that generated sound arrives as material rather than as a sealed object. Recordings remain precious, and sampling culture is built on them. But a recording is something you listen to. A project is something you continue. Programming learned that distinction first.
Programming went first
Software development has already been through the transition music is entering. For decades, code was written in an editor, built by a compiler and explained by a manual, three tools learned separately. Now a developer states an objective; an assistant changes the files directly and runs the tests; version control records every change with its author; the developer keeps or reverts. In Stack Overflow’s 2025 survey, 84% of developers were using or planning to use AI tools, and about half of professional developers used them every day.
What made this work was not the chat window. It was four quieter things: a shared editable object (the repository), the same commands for the person and the agent (the agent edits files like any other editor), senses (compilers, tests and logs that tell the agent what actually happened), and memory (version control, so nothing is lost and every change has a name). “Cursor for music”, after the AI code editor, is shorthand for the same four things in a song. Bring the bass back four bars later and leave the vocal alone: the agent makes that change in the song itself, and the person hears it in context and keeps or reverts it.
The analogy breaks at one point, and it is the most important point in this essay. Tests can check code. Nothing can check taste. A measurement can say whether a kick lands on the beat, whether a mix clips, or whether a tail decays. It cannot say whether a phrase says what you meant. In music, the test suite is a human ear. Everything the next DAW must do follows from that.
What the next DAW must do
Five properties follow, each a relationship between a person, an agent and the music rather than a feature.
Start from the intention, at every depth
People think about music in musical terms: hear the whole, attend to a part, change a sound, hear the result in context. The instrument should let them work at whatever depth the intention lives (“let the chorus arrive later” is arrangement; “a warmer bass” is sound) and at whatever depth the material honestly supports; a recording does not automatically yield clean parts, and the instrument should say so rather than pretend.
A first useful action should be reachable without learning the whole tool: a beginner may stop after one change, while an expert may need one note in one bar, and the same work serves both. A low floor, never a low ceiling.
Let people and agents edit the same song
In the next DAW, an agent is not a separate app that sends back a file. It edits the song directly, like any other editor, within the access the person grants, whether that is one change or a whole task such as a kit for a fast, hard track. Every operation the agent performs is one the person could perform by hand, so the agent’s work can also teach.
There is no private project that only the agent can see, no change that exists in the conversation but not in the music, and no second door: no approval queue that an agent’s change must pass and the person’s own change need not. Close the conversation, and every change, every version and every control stays where it was.
Hear for the agent
An assistant that cannot perceive the music will invent it: a confident description that says nothing true, a key guessed from a filename, a layer that fights the kick. Agents do not hear, so the instrument must hear for them.
That means typed musical evidence produced by the instrument itself, not imagined by a model: the key as a ranked distribution rather than one guessed label, beat grids with their confidence, energy and brightness over time, and whether a new part would fill an open pocket or clash with what is there. When the evidence is ambiguous, the agent should receive the ambiguity. Two different assistants on two different days should read the same truth about the same audio. The agent’s judgment may vary; the evidence must not.
Make every sound finished and editable
A producer who wants an industrial bass for a hard techno track wants that bass, finished. Samples will keep that role. What changes is what stands behind the sample: when a sound carries its editable source, it can be re-keyed, re-timed or re-shaped when the song’s needs change, instead of being searched for again. The sample, the preset and the lesson become one object: a finished sound, the settings that make it, and an explanation an agent can give.
The cleanest way there is one generator for every native sound, so that a kick, a clap, a bass and a pad are regions of one space rather than four separate synthesizers. A clap can start from a kick’s own settings; a groove keeps its timing when the kick is swapped; and text about each sound, its settings in human units, its neighbours and its measured render, gives an agent eyes and ears in the space, including the honesty to say when it does not know. Whether one space can match instruments built for a single kind of sound is a question the ear must settle, kind by kind.
Remember the work
Programmers take version control for granted; musicians mostly do without it. The next DAW should record every change with its maker, person, collaborator or agent, and let anyone allowed to edit keep a version, the way a commit keeps code. Any version can be heard, compared with any other and returned to; nothing kept is lost by going back.
Memory makes experimentation safe, because the ending you had is never at risk while an agent tries three others, and it makes music continuable. Imagine publishing a piece with its arrangement and its sounds. Someone hears it, enters the drum part, tries a different rhythm and keeps a descendant. The original stays intact, and its maker can hear the answer and continue from it in turn. A work becomes an invitation to a musical answer, and a good answer becomes another beginning. Whether people want that often enough to sustain it is a hypothesis, and sharing stays an act the person chooses. So does the last word.
The verdict stays with the ear
Who is the author of a song made with an agent? The person who directs the work, who writes the brief, guides each pass and decides what stays. The record keeps what an agent or a generator contributed distinguishable, so the history tells the truth about how the music was made. Directing an agent is a form of expression, not the absence of one. Music is music, whether made by hand, with an agent or in any tool; what matters is access to its building blocks, so that it can be heard, changed and built on.
What must never move is the verdict. Evidence is what the instrument measures; the verdict is what the ear decides, and no reading can stand in for it. A loudness, a pitch and a brightness describe a sound; whether it is a kick you would use is your answer, given after listening. An agent may judge its own attempts and say which it set aside. It may not claim to have heard the music, and it may not put its choice in place of yours.
That is the whole thesis. Here is what would break it.
What could prove this wrong
It is a thesis, not a prophecy, and each objection to it can be tested.
Most people may simply want to listen. If people offered an easy way to change a song mostly go on listening, the demand for shaping is a minority one, and the instrument must be honest about that.
Participation can be attractive and still lose. Player-piano sales in America peaked in 1924; within a few years the electrical phonograph and radio had swept them aside, and the crash of 1929 ended their prominence. Many buyers seem to have valued the outcome more than the act of steering. The question for any new instrument is whether people return for the act itself, or only because it is, for now, the easiest way to a result.
The incumbents can absorb it. Free, integrated studios already reach enormous numbers of people, and the established DAWs can bring assistants and sample libraries inside. A new instrument must be clearly better at something people do often.
Review may cost more than it saves. An agent that works while you are away can leave you with more changes than you can hear. If people stop listening to what agents change, the agents’ reach must narrow.
Recorded character may win. A recording carries performance, room and history that synthesis may never reach. Offered an editable and a frozen version of the same part, producers may choose the frozen one. Then recordings must stand beside editable sounds as first-class material.
Consent and displacement are unresolved. New instruments have always met organized objection from the players they displaced; synthesizers did too. Assistance that can make whole songs raises the question again, together with the consent of the musicians whose recordings trained the models. In June 2026 the American Federation of Musicians sued Universal and Warner, alleging that recordings made by its members were licensed to Suno and Udio without compensation or credit. These questions do not have engineering answers.
The order of change
Nor does the change arrive in the order one would design. The intuitive order is:
- first, a new kind of DAW;
- then, new habits around it.
The likelier order is the reverse:
- first, agents arrive inside the work people already have, including inside the workstations they already use, bringing better senses than those workstations have;
- then the editable project, not the finished recording, becomes the object people and agents share, keep and continue;
- and only later, if at all, a new instrument replaces the studio metaphor.
Software followed the same sequence: assistants entered the editors developers already used well before anyone argued for a new kind of editor.
The studio metaphor will not end the day an alternative becomes better at everything. It will begin to end the day an alternative becomes better at the thing people do most: changing a song and coming back to it.
The instrument of the past and the future
This is the instrument I am building. It is called Dombra, after the two-stringed lute of the Kazakh steppe, a living instrument that has carried songs and stories for centuries. The name holds the ambition: an instrument of the past and the future, in which a gesture becomes sound now and the work survives to be continued.
The mark is two strings across three bands of colour: red for the lows, green for the mids, violet for the highs. The low band is longest because low frequencies carry furthest, which is why the kick is what you still hear outside a club. The colours are an analogy, not a mapping from sound to light.
Dombra’s musical agent is called Seyre, after a singing bird. Seyre generates sounds and melodies and guides you through the musical process. Ask it for a kit, a groove, a hook, a bass line or the next section of a song, and it builds the thing in the song itself, through the operations your own hands would use; its own model makes the material a sound or a section needs. Every step is an ordinary, recorded change you can hear, step through, undo or keep, so working with Seyre also teaches you the moves. It reports what it changed and what it could not do, and it carries you to the next decision you can hear rather than making it for you. It does not compete with the general agents you already use; it works with them, and gives them the instrument to work through. You direct, and you keep the verdict.
Every decision in Dombra answers one question: how does this improve human expression, or help someone realize their musical intention? Whose intention, what becomes better, what stays under their judgment, and what evidence would show that it worked.
The DAW of the past was a studio you had to learn before it would serve you. The DAW of the future meets you at your intention, lets an agent carry the work you do not want to carry, keeps every choice you might want back, and leaves the last word to your ear.
The studio is still standing. But the first versions of its successor are already on the workbench.