The debate around AI music often starts with generation. Can a model make a convincing song? Can it produce a believable vocal? Can it recreate the sound of a band?
Our recent conversations with artists suggest that these questions are becoming less useful. The more interesting question is what happens to the roles surrounding music when generation becomes easier.
In The Story Behind AI Music Artists You Might Have Missed, Doniell Clay showed how artificial performers can become vessels for human-written stories and carefully constructed identities. In AI Music and The New Burden of Choice, Vitalii Klimenko argued that cheap generation increases the importance of production decisions. Brian Myk then showed in What Skills Should Artists Value in the Age of AI Music? how the bottleneck can move from production logistics toward judgment.
Our conversation with Stu Bedford adds another perspective. Bedford is primarily a songwriter. He has spent years writing music, but performing, producing and assembling musicians were never the parts of the process he was most interested in.
Through his project An Act of Treason An Act of Treason on SoundCloud, he is now turning songs he has written into finished records using AI-generated performances and production. His experience raises a larger question. If convincing performance becomes easier to generate, could the ability to write a meaningful song become more important?
From a song in your head to a finished record
Bedford's introduction to AI music began with a practical problem. He had written a ska and pop-punk song called "Weeds" and imagined it with a lively horn section, but his stock horn sounds in Logic were not producing the result he wanted. He did not play horns himself, and his own production skills were becoming the limitation.
He had encountered the traditional solution before. For an earlier studio album, Bedford had hired a horn player for another song. Bedford gave him the direction, and the musician supplied the expertise. The result worked, but it also cost money and took time.
Generative AI changes the economics of exactly this type of decision. After struggling with "Weeds", a songwriter friend suggested Suno. Bedford initially resisted the idea because he associated AI music with entering a prompt and receiving an automatically written song.
Instead, he uploaded music he had already written. The result gave him something much closer to the production he had imagined, including the horn section he had been unable to create himself.
That changed his understanding of the technology.
"The distinction for me is that I write the songs; AI helps me turn them into records."
Bedford brings his chords, lyrics, melodies and structures. The model helps him explore what those songs might sound like as finished productions. For a songwriter with hundreds of unfinished originals, that difference is substantial.
Producing all of them traditionally would require considerable time and money. AI does not necessarily give him more ideas to write. It gives more of the ideas he has already written a route toward becoming records.
There is an uncomfortable economic side to that freedom
Bedford's horn example also reveals one of the more difficult consequences of generative production. Previously, there were musical tasks he could not perform himself, so someone else could be paid to perform them. Today, at least some of those tasks can be generated.
This is not a literal story of Bedford replacing the same horn player with Suno. The musician was hired for a different recording. But the comparison shows how the economics can change.
A songwriter who wants a horn section once had several options. They could learn the instrument, settle for software instruments, find a collaborator or pay a musician. Generative technology adds another option that can be dramatically cheaper and faster.
For the songwriter, that is access. For the session musician who might otherwise have received the work, it can mean lost demand. Both can be true at the same time.
AI music discussions sometimes struggle with this tension because the benefits and costs fall on different people. Bedford can now realise songs that might otherwise remain demos, which is creatively valuable. Yet if thousands of creators make the same decision, some production work that previously flowed toward singers, instrumentalists and session musicians may no longer do so.
That does not mean human performance becomes obsolete. Live music, distinctive performers, specialist musicianship and human collaboration still offer things a generated recording does not. Bedford himself describes his traditional studio collaboration positively, but generative tools can change which projects justify paying for those skills.
AI may be particularly important for songwriters
Bedford is also interesting because he does not fit the traditional image of someone desperate to become the person standing at the front of the stage. He told us directly that he dislikes performing and is far more comfortable writing songs in his spare bedroom with a guitar.
Production is another specialist discipline he tried to learn, but the time he could have spent improving as a producer kept competing with the thing he actually wanted to do. Writing was where he wanted to spend his creative energy.
This distinction has existed throughout music history. Songwriters have long written material for other performers, while producers, arrangers, session musicians and vocalists help turn those compositions into recordings. Generative AI changes the number of people who may be required between the songwriter and the finished record.
Bedford described a previous studio album that cost around 15 grand and took two years to make. Ten songs reached completion. He enjoyed the experience, but that production model could never keep pace with how much music he wanted to write.
AI gives someone in his position a different route, and that may prove to be one of generative music's most consequential effects. There are likely many people who love composing, writing lyrics or developing musical ideas but have little interest in performing.
Others may lack a strong singing voice, instrumental ability, production knowledge, money or access to collaborators. Historically, those limitations could determine whether their songs ever became more than demos. AI weakens that connection.
Does songwriting become more valuable when performance becomes cheaper?
Bedford thinks it might. His reasoning is straightforward. If a polished performance becomes easier to produce, he becomes more interested in what the song itself has to say.
For his upcoming album "Solo Levelling", Bedford says AI generated the vocal and instrumental performances. But the material underneath them is highly personal, covering relationships, mental health and his own experiences.
That creates an important distinction between generating performance and determining what that performance is supposed to communicate. As Bedford points out, the technology did not live the experiences behind the album or decide what he needed to say about them.
The easier technically competent music becomes to produce, the less technical competence alone can distinguish a song. A clean vocal is less remarkable when thousands of people can generate one, while a large arrangement is less distinctive when another can be created in minutes.
Production quality still matters, but abundance can reduce its ability to function as the sole differentiator. That may push attention back toward songwriting, concept, perspective and emotional specificity.
Bedford uses his own writing as the standard against which generated results are judged. Does the vocal carry the intended feeling? Does the arrangement leave enough room for the lyric? Does the production sound impressive, or does it actually sound like his song?
Those are not simply questions about whether the model succeeded technically. They are questions about whether the technology served the writing.
AI as a collaborator rather than a vending machine
Bedford also makes an interesting comparison between generative AI and his previous studio experience. His earlier song "Katie" began as a fast punk song inspired by The Offspring. During production, he proposed turning it into what he describes as a "punk rock western".
A producer then helped transform that idea through fiddle, guitars and other production choices. The finished record was not created by Bedford alone, even though the song and initial direction came from him.
He sees something familiar in AI-assisted production. He brings the song and an intended direction, something comes back, and he listens to it, rejects parts, follows unexpected ideas and decides whether the result still serves the song.
The major difference is speed. Bedford says a transformation that once unfolded across months can now sometimes happen in one or two hours, including listening, rejecting ideas, experimenting and deciding when the record is finished.
This reinforces something we have heard repeatedly in our previous interviews. AI does not simply make one production process faster. It can make the process more fluid by allowing more possibilities to be tested before commitment.
That increases access, but it also increases the number of decisions the creator must make. Generation becomes easier while judgment becomes a larger part of the process.
The production bottleneck falls, the attention bottleneck remains
There is another part of Bedford's experience that deserves just as much attention as the production technology. AI has helped him finish more music, but it has not solved the problem of getting anyone to listen.
Bedford's previous studio album received significant investment, yet he describes it as sitting in what he calls "Spotify purgatory". The production was completed and distribution existed, but there was not enough marketing or audience development to move the record further.
His conclusion is simple.
"Getting the music heard is always going to be a thing."
This may become even more important as AI lowers production barriers. If considerably more people can make polished music, considerably more polished music competes for the same listeners.
The scarcity does not disappear. It moves. Studio time becomes less scarce, basic production becomes less scarce and convincing vocals may become less scarce, but listener attention does not.
This means marketing, audience building, artist identity and distribution may become increasingly important parts of releasing music as production itself becomes more accessible. The irony is that AI can solve the problem of making a songwriter's work ready to be heard while making the competition to actually be heard even more intense.
Who is the artist if nobody performed?
This leads naturally to Bedford's final argument. Music already separates writing from performance. Major performers routinely sing songs written partly or entirely by other people, yet we still recognise the performer as an artist because performance, interpretation and identity carry their own creative value.
Bedford asks whether AI requires us to examine the relationship from the opposite direction. Can someone be the artist behind a record without being the person singing or playing it? For him, the answer is yes.
He writes the lyrics, creates the vocal melody, chords and structure, and decides how the finished work should feel. He evaluates the generated performances and decides what belongs, but he does not claim this makes him the instrumentalists or vocalist we hear on the recording.
AI-assisted authorship does not require pretending that generated performance is human performance, but the absence of physical performance does not necessarily remove every other form of artistry either.
Bedford's framing may therefore point toward a broader change in what the word "artist" means. The traditional package of songwriter, performer, instrumentalist and public personality may become easier to separate, allowing someone to primarily be a writer and creative director while someone else remains primarily a live performer.
Another artist may combine traditional instruments with generated production. AI is unlikely to force all of those roles into one new model. It may instead make more combinations viable.
AI music is revealing how many jobs were hidden inside the word artist
Across our recent interviews, a pattern is becoming clearer. Doniell Clay showed how fictional performers can carry the perspective and experiences of a human creator, while Vitalii Klimenko showed how generation can turn production into a larger process of exploration and decision-making.
Brian Myk then showed how cheap alternatives can make judgment, taste and rejection more important. Stu Bedford adds another layer, showing how AI can allow the songwriter to move much further through the production chain without becoming the singer, instrumentalist or producer they previously needed around them.
That creates opportunity, but it also creates displacement. It makes old songs economically feasible to finish while potentially reducing demand for some of the people who previously helped finish them, and it does nothing to guarantee that anyone will listen once the record exists.
Those contradictions are likely to define much of AI music's next phase. Perhaps the biggest mistake is assuming that AI simply makes "the artist" more powerful when what it really does is expose how many different activities were bundled into that word in the first place.
Writing, performing, producing, arranging, selecting, marketing and building an audience have often existed under the same broad idea of making music. AI can redistribute those responsibilities without making all of them disappear.
For Bedford, the important part is clear. He writes songs because he feels compelled to write them, and generative technology means more of those songs can now become finished records even if he never wanted to stand on a stage and perform them himself.
That possibility may eventually matter far beyond AI music. It suggests a future where being able to write a record and being able to perform a record become increasingly separate roles, and where the question of who counts as the artist depends less on who physically created every sound and more on who gave the work its reason to exist.
