I was working on an idea inspired by the Japanese composer Yasunao Tone, who is known for his work with damaged CDs, deliberately scratching and altering them to produce unpredictable digital glitches. Instead of physically modifying a medium, I attempted a digital equivalent—randomly changing the binary data of a MP3 file while keeping the header intact. The acoustic results were reminiscent of early digital noise experiments: chaotic yet oddly predictable, lacking the dynamism or depth that might make them feel agentic or expressive. The outcome, despite its apparent randomness, felt aesthetically dull.

One aspect that stood out was the relationship between agency and the process itself. Initially, I altered the binary code manually, making direct changes to the file without any real craft or skill—simply flipping bits at random. Later, I wrote a Python script to automate the process, generating variations instantly with minimal effort. Curiously, despite the manual approach requiring no particular craftsmanship and the programming being the more technically demanding task, the act of making changes by hand felt more “hands-on,” more like an act of direct intervention. The immediacy of automation, by contrast, distanced me from the process, reducing my role to that of a passive observer of outcomes rather than an active participant. This tension—between procedural mediation and perceived agency—became central to my reflections on the experiment.

Even at this early stage of research, the problem of agency becomes pressing. Does any form of mediation represent a loss of agency? How many of these losses are known and visible to me? Where does agency begin, and where does it go once it moves? Is it something held, or does it pass through bodies, infrastructures, codes—always contingent, always in transit? Agency is often treated as a possession, something one has or lacks, as if it could be accounted for like a commodity, like a resource that is depleted when shared or contested. But what if agency is something that circulates, a movement rather than a substance, a process rather than a property?

To ask this is already to displace the question from a static ontology toward something more fluid, more topological. One does not simply lose or gain agency but finds oneself positioned within channels that accelerate, redirect, or obstruct its movement. Bureaucratic structures, for instance, do not necessarily suppress autonomy but reroute it: decisions are not eliminated but proceduralized, dispersed across layers of regulation, embedded into the slow metabolism of institutions. Who, then, is the agent? The individual who initiates an action, or the system that conditions its trajectory? Is control to be found in the formal structures of decision-making, or in the frictions and delays that structure participation?

Perhaps a hydrodynamic model offers something here, a way of seeing how agency pools, gathers, is dammed, is diverted. Just as water follows the contours of a landscape, agency seems to follow the shape of systems—accumulating in institutions, in capital, in technologies that have been historically advantaged. Digital platforms, for instance, do not merely host engagement; they channel it, shaping its flow, drawing energy from the currents they direct. A user produces, interacts, participates—but always within the infrastructural logic of the platform, where every gesture is transformed into data, every act of engagement absorbed into an economy of extraction. To what extent does this capture reconfigure what it even means to act? And yet, even within these constraints, counterflows persist—distributed networks, lateral resistances, movements that seep into the cracks of dominant structures, operating not through accumulation but through diffusion.

But if agency is a current, what happens when it is no longer visible? If power is not merely a thing possessed but a pattern of circulation, then its stakes lie in the infrastructures that make movement perceptible—or render it opaque. This is the question of governance, of the architectures through which agency is made legible or imperceptible. Algorithms distribute agency in ways that resist observation, shaping decisions through imperceptible nudges, redistributing power through logics that rarely announce themselves. Can one act freely if the pathways of action are pre-filtered, pre-structured, imperceptibly weighted toward certain outcomes? And how does one resist a system whose operations are hidden, whose sluices remain invisible even as they direct the flow?

Composition has always been mediated by its tools; notation is not merely a record of music but a mode of thought, a way of organizing sonic possibility. How much, then, does agency reside in the composer, and how much in the grid of the score, the formalized constraints of legibility that determine what is writable, what is playable, what is thinkable?

What shifts when notation moves from paper to software? In programs like Sibelius or Finale, the act of composition is no longer just inscription but navigation—through menus, through predefined instrumental palettes, through default quantizations that smooth rhythmic complexity into conventional grids. One does not simply write music in Sibelius; one interfaces with a system that suggests, auto-corrects, aligns, anticipates. What is made easier? What becomes more difficult, or disappears altogether? Is the composer’s agency amplified by these affordances, or does it find itself gradually absorbed into the software’s logic?

And what, then, of AI? If notation software gently prefigures the act of composition, AI-assisted composition marks an intensification of this mediation, moving from a system that facilitates notation to one that proposes material, generates structures, simulates stylistic continuities. The question of agency here becomes even more elusive: Is the composer steering the machine, or is the machine subtly steering the composer?

Perhaps, then, the loss of agency is not the central concern. Perhaps the more urgent question is: what kind of agency emerges in these new configurations? Does AI composition distribute agency in ways that exceed the human composer, opening new potentials, new sonic landscapes? Or does it deepen existing sedimentations, reinforcing the styles, forms, and harmonies that the training data privileges, subtly guiding musical thought into well-worn channels? What would it mean to compose against the current of AI, to carve new tributaries, to force a rupture in its predictive logics? I tap again on the idea of composing in glitches.

To think agency as fluid is to think not in terms of possession but in terms of mapping—tracing how agency is routed, how it is structured, how it might be redirected. The challenge is not merely to ask who has agency but to investigate where it gathers, how it moves, what barriers channel or impede its course. How might one intervene in these structures, not to reclaim a lost sovereignty but to reconfigure the conditions of possibility?