The most useful way to understand generative AI in music may not be as a replacement for composers. In many serious workflows, it is becoming something more specific and more disruptive. It changes how quickly a creator can explore, compare and refine possible versions of an idea.
Klimenko VG, independent composer and artist and founder of Binary Area Records, offers a clear case study. His compositions, lyrics, melodic ideas and dramatic intentions begin with him. Generative systems then enter downstream as production instruments, allowing him to hear many possible interpretations of a musical idea that already has an authorial direction. His summary is precise. “Generation creates options. Production creates decisions.”
That sentence matters because it separates two activities that are often collapsed in public arguments about AI music. Generating a result is not the same as making a record. A generated output can be impressive, useful, wrong, misleading, too generic or strangely close to the emotional centre of a song. The production question is not only whether the system can create sound. It is whether a human producer can recognise which sound serves the work.
A useful comparison is Grammarly. The writing tool can correct, suggest and refine, while the essay still carries the writer’s argument and intention. Generative music can function in a similar layer of the process, although its influence can be much larger. It does not merely smooth a sentence or fix punctuation. It can offer alternative sonic interpretations, performance characteristics, production textures or mix directions that change how the underlying composition is perceived.
That makes the analogy helpful but incomplete. In writing, a suggestion may sharpen an argument. In music production, an alternative version may reveal that a song wants a different tempo, a different tonal colour or a more restrained arrangement. It may also seduce the producer with polish while weakening the composition’s identity. The machine may generate fluency. The producer still has to decide whether fluency is enough.
Klimenko says he has explored more than 20,000 generated outputs across months of work. That number should be treated as practitioner experience, not a scientific benchmark. The important point is not the exact count. It is what happens to the bottleneck when another alternative becomes cheap.
When generating one more version costs little time and little money, generation stops being the difficult part. Selection becomes difficult. A producer can test directions that would previously have been abandoned because recording them required musicians, studio time or substantial production effort. A sparse piano interpretation, a heavier electronic treatment, a cinematic build or a different vocal character can be auditioned before the producer commits.
This shifts pressure onto taste. One generation may be more polished. Another may contain a better musical idea. A spectacular result may still need to be rejected because it changes the identity of the composition. That last point is essential. The best sounding output is not automatically the right output. Production has always involved refusal, but AI makes refusal more visible because the pile of plausible alternatives grows so quickly.
From creation to investigation
Klimenko’s most interesting idea is that AI accelerates not merely music creation, but musical investigation. A thought can become an experiment almost immediately. That experiment can reveal another direction, which can itself be tested. The process becomes less linear and more recursive. An idea leads to an experiment. The experiment produces an alternative. The alternative demands comparison. Comparison produces rejection. Rejection leads to a new experiment. Eventually, a decision has to be made.
This is not a minor workflow change. Historically, producers were constrained partly by the cost of testing ideas. Even in home studios, time, skill and available performers shaped what could reasonably be tried. Generative systems lower that cost, which expands the field of possible versions around a song. For creators, this can be liberating because weak assumptions can be challenged quickly. For listeners, it may mean finished tracks that have passed through more variations before release. For platforms and AI artist ecosystems, it raises a different question. How should we evaluate work when the act of choosing among generated possibilities becomes a central creative act?
At Lunar Boom, we see this distinction as increasingly important because AI music is not one workflow. Some users may ask for a complete track from a prompt. Others may compose first, then use generation to test arrangement and production paths. Others may combine editing, stems, human performance and generated material. These workflows carry different creative responsibilities, even when the final listener only hears a finished song.
Abundance creates its own problem. If another version is always one click away, creators can enter an endless loop of generation. The important question becomes not “Can I generate something else?” but “Why am I generating another version?” Without that question, exploration becomes avoidance. A producer can mistake motion for progress, collecting alternatives instead of making choices.
This challenges the assumption that more automation necessarily means less creative responsibility. In some workflows, more possibilities require more judgment. The producer is not only operating a tool. The producer is defining the identity of the work by deciding what belongs, what distracts and what must be discarded. Editing, curation and rejection become creative acts rather than administrative chores.
That also changes how production skill may be understood. If more creators eventually gain access to systems capable of generating technically competent music, access to generation itself will provide little differentiation. Value may shift toward taste, songwriting, editing, curation, arrangement, rejection, creative direction and the ability to recognise what actually makes a song good.
This has implications beyond individual producers. Musicians may increasingly need clearer language for distinguishing composition, performance, production, generation and creative direction. Artists working with generative systems will need coherent creative direction if they are to become more than a stream of outputs. Rights holders and platforms may need clearer language for describing how works were made, especially when the difference between composition, generation and production affects how people understand authorship and trust.
The broader implication is simple but demanding. Generative AI lowers the cost of trying an idea, but it does not remove the need to know which idea deserves commitment. The future producer may spend less time asking “How can I make this sound?” and more time asking “Of all the ways this could sound, which one deserves to exist?”
That may be one of AI music’s biggest changes to music production. Not the removal of human decisions, but the rising importance of their quality.
