Practicing Scales

The album I couldn't have made

The songs began as poems, written during a real period of my life: persistence, rejection, vulnerability, resilience. I fed them to Suno, an AI music generator, and shaped what came back: rejecting versions, changing keys, adjusting pacing and instrumentation, pushing until each track felt faithful to what the poem originally meant. My role wasn't performer. It was something closer to producer, or creative director.

The result was a finished album that would have been flatly impossible for me alone. I felt my creative capacity expand.

Then a friend, trained in theatre, heard it and recoiled. To her, AI-generated work diminished the value of every artist who spent decades earning their craft.

I've come to think we were both right. Untangling how we could both be right requires better questions than “is AI good or bad for us?” It requires asking what any learning process actually builds, and what happens to that building when a machine takes over the work.

Every new medium made the same promise

As a part of my Master's at HGSE, I am researching how past technologies were adopted into the classroom. The historian Larry Cuban traced a century of technologies that arrived promising to transform school, and mostly didn't.

Neil Postman argued that a medium is never neutral; it contains an epistemology, quietly reshaping what we count as knowledge and what kinds of thinking we value. Paul Saettler showed that every educational technology carries an implicit theory of the learner: a receiver of information, a behavior to be shaped, an information processor, a constructor of knowledge.

So the real question about AI isn't what it can do for my learning. It's what it believes I am trying to become. And to answer that, I had to look at what my own learning had actually built.

What ten years of piano was actually for

I studied piano for more than a decade as a child, mostly reluctantly. Scales, exams, repeating pieces long after they had gone boring. For years I understood that experience as one thing: learning music.

Looking back, the performance was the smallest part of it. The learning process formed capacities that were invisible in the final performance. I realized this years later as an adult after watching my mom, who decided to pick up the piano in her 50s, play at a recital where the other performers were kids. It was never really only about the music.

Where the work went

AI didn't simply “help me make music.” Every piece of the work went somewhere, and the destinations are not equal.

• Preserved with me: the lived experience behind every song, the poems themselves, selection of poems into songs, and final say on every track.
• Offloaded: arranging instruments to the melody I created on Voice Memo, vocal performance, and production and engineering.
• Capabilities I did not build: I did not become a better singer, learn to arrange, or learn to engineer a recording.
• Capabilities that expanded or relocated: articulating exactly what I want, comparing many alternatives quickly, and directing a creative system.

Some agency moved up the stack, from executing to directing. Some was preserved. And some developmental pathways were quietly closed, because the work that would have built them was never done by me.

So did AI erode my agency?

This turns out to be three separate questions wearing one coat. They can move in different directions at the same time, and for my album case study they did.

People who don't train for ten years can make AI music. But this is the trap in the popular claim that “AI moves humans from execution to judgment.” Judgment has prerequisites. Without enough domain knowledge, what looks like judgment is really just preference. I could meaningfully offload execution because I had once done the execution the slow way.

The goal isn't preserving struggle for its own sake; plenty of friction is pointless and technology should remove it. The question is: which experiences are load-bearing for human development, and where does the difficulty go when the machine steps in?

My gain, whose loss?

The economists Amartya Sen and Martha Nussbaum argue that education shouldn't be judged only by its outputs but by whether it expands people's real capabilities: to reason, create, choose, and pursue lives they have reason to value. By that standard, my album passes. My capability to express something true about my life genuinely expanded.

But my theatre friend wasn't wrong either. Anyone with something to say gets access to the fruits of expertise without the decade of training. Expression is democratized. At the same time, if the outputs of craft are free, the incentive to spend ten years acquiring the craft weakens, along with the ecosystems that train, employ, and sustain the people who have it.

A tool for auditing your own AI usage

The interactive essay compresses the argument into a self-audit built around three dimensions of AI-mediated agency. It asks the reader to consider a task regularly handed to AI and distinguish between the ability to execute, the ability to judge in the field's own terms, and ownership of the meaning or purpose of the work.

The audit describes four zones: AI slop, vibe curator, hired gun, and director's chair. It also adds a “Skipped practice” modifier, asking whether the judgment a person relies on was built before AI arrived, and who will do the repetitions that build future judgment.

The artifact and the person

AI can produce an impressive artifact without producing an equally capable person. It can also let a capable person express, create, and act at a level that was previously out of reach.

The difference is which scales we've already practiced, and which ones we are skipping.

Using this audit tool, my album may currently sit in the “Director's Chair” zone, but I know it only got there on a decade of scales I practiced before AI existed. That's the question worthy of asking of every AI-mediated learning experience: not what did the technology let the learner produce, but what work was offloaded, what would that work have built, and which of those capacities were lost, preserved, or relocated into new forms of agency?