Composing the Conditions

Working decision

Working title: Composing the Conditions: How Generative Functional-Music Apps Reshape Digital Musicianship

Use Endel as the sustained case within the wider field of generative functional-music apps. Its normal soundscapes combine a function-first interface, authored sound material, runtime generation and contextual inputs. The Grimes AI Lullaby collaboration gives the case one artist example without turning the talk into a presentation about Endel or Grimes.

Research question

When musical form is realized at runtime, what becomes the work, who makes it, and who controls it?

A fuller academic version is: How do generative functional-music apps redistribute musicianship when creators design materials and constraints, software realizes one sequence, listeners and context supply conditions, and a proprietary service controls access and change?

Main argument

Generative functional-music apps move some decisions about sequence, duration, density and transition from a fixed recording into runtime. Creators design sound families, constraints and mappings, then judge how the system behaves across many sessions. Musicianship remains, although part of it moves from fixing each event toward system design, curation and behavioral listening.

The change has three parts:

1. Musical decisions move downstream: sequence, density, duration, transition or processing can remain open until playback.
2. Musicianship moves upstream: skill shifts toward making reusable material, defining invariants and variables, mapping inputs, testing many realizations and judging long-term system behavior.
3. Control remains unequal: artists, developers, interfaces, listeners and context can affect what sounds, while the service owner controls functions, success criteria, runtime updates, access and the data relation.

The short answer: creators make materials and constraints, the runtime produces one realization, listeners and context condition it, and the service owner governs access and change.

Revised second-half argument

The post-market section now follows one distinction: contribution to a heard session is not the same as control of the system.

1. Affordances and automation: The function button gives the listener coarse control over intended use and turns that use into a production criterion. Predesigned material and logic bound the session while some choices remain open until playback.
2. Digital musicianship: Work includes modular composition, constraint design, long-duration listening, failure finding and maintenance. This remains musicianship even when no one fixes the final sequence note by note.
3. Distributed authorship: Artist, sound team, developers, software, listener and context can affect one realization. Their causal roles do not give them equal authorship, ownership or power.
4. Functional listening and platformization: A function label moves from classifying finished tracks to constraining current musical form. The interface frames a desired state while listeners may still bring taste, memory, pleasure and judgment.
5. Generative AI: Endel’s public architecture uses predesigned elements and logic, with an unclear machine-learning role. Current live music models can synthesize continuous audio under changing controls. They extend the runtime production layer rather than inventing the overall format.

The critical claim is narrow: generative functional-audio apps make musical form variable inside a service whose categories, data relations, updates and access remain centrally managed.

Lecture theory audit on 22 July 2026

The current second half identifies the right concepts, yet it often treats them as separate labels. The lecture slides support a clearer chain: a function label defines a desired state, musicians translate that state into musical behaviour, software realizes one session, listener and sensor data modulate it, and the service owner controls the system in which all of this occurs.

The service as the unit of analysis

Medium theory separates the micro level of tools and interfaces from the macro level of infrastructure, labour, genre and value (Session 1, p. 22). Endel connects those levels in one service:

• The listener interface offers functions instead of tracks.
• The production system stores material, mappings and limits.
• The runtime turns those conditions into one session.
• The company controls access, data relations and later changes.

The algorithm alone cannot explain the musical work. The proprietary service acts as musical medium, production environment, playback system and governing institution.

Affordances beyond technical limits

The course distinguishes compositional, behavioural, embodied, listening, aesthetic and temporal affordances (Session 3, p. 63). Endel makes each category concrete:

• Compositional and temporal: endless playback, modular material, weak boundaries and gradual change.
• Behavioural and listening: a listener chooses a purpose, then gives up most control over sequence and musical detail.
• Aesthetic: focus and sleep favour continuity, restrained dynamics and low interruption.
• Embodied: movement or heart rate can affect sound as data rather than deliberate musical gesture.

The embodied case deserves its own question. A body can change the session without intentionally performing. This makes the listener causally present without automatically making them a musician or author.

Automation as stored musical judgment

The sequencer lectures describe automation as an inscription of performance. Repetition, timing and modulation move into a running system, so the phrase becomes a loop, performance becomes programming and improvisation becomes parameter change (Session 3, pp. 23, 25–26).

Endel continues this history at service scale. Musicians and developers store judgments about acceptable density, transition, repetition and change before playback. The runtime performs within that field. Automation therefore relocates decisions across time instead of removing them.

Musicianship as translation

The course defines interface literacy as the ability to read musical codes, select material and organize it within software (Session 5, p. 31). Its AI lectures extend this into semiotic labour: musicians guide systems through labels, prompts and metadata rather than specifying each event (Session 12, pp. 5–6, 65; Session 13, pp. 63–64).

For functional audio, the central act is translation. A production team turns words such as Focus, Sleep and Relax into sonic rules. This work includes:

• deciding what the function should sound like
• selecting sounds that carry the intended meaning
• encoding acceptable variation and failure
• judging sessions against a product goal

The current deck explains modular composition and testing, but it should name this translation as musicianship. The musician does not only make assets and rules. They define a musical version of a desired human state.

Several times of composition

The course treats modular production as intermittent authorship: contributors act at different points, and their work may circulate without stable visibility (Sessions 9, p. 35; 10–11, pp. 39, 42). Endel divides composition across time:

1. Artists or sound designers make source material.
2. The production team defines behaviour and limits.
3. Software, function choice and permitted context produce one session.
4. Later updates can change future sessions.

This is more exact than saying that authorship is simply distributed. It also raises a question about identity: if the company changes the mappings after release, does the same musical work still exist?

Participation without equal control

The platform lectures distinguish participation from power. Users navigate, choose, listen and produce data inside options defined by the platform (Sessions 10–11, pp. 79–82). Endel makes that difference audible because listener data may affect the current session.

The listener contributes purpose, context and attention, yet the service defines the available purposes and possible musical responses. This supports a sharper distinction among causal input, authorship, ownership and governance. Distributed causation can coexist with concentrated control.

Function as musical value

The functional-listening slides define value through support for tasks, state regulation and integration into daily routines (Sessions 10–11, pp. 89–91). This changes the criteria by which musicians and services judge music:

• low interruption can count as success
• fatigue and distraction become production failures
• continuity can matter more than memorable form
• efficacy claims can carry more weight than artist identity

This is stronger than a choice between functional and aesthetic listening. The critical question is which musical qualities become valuable when the product promises an effect, and who gets to set that standard.

Function labels as primitive prompts

The AI lectures describe a move from direct execution toward guiding systems through signs, language and metadata (Session 12, pp. 5–6, 65). A function button already follows this structure. The listener states an aim in words, and the system translates that aim into sound.

AI can expand the material generated inside this relation, but it does not invent the relation itself. The stronger outlook is therefore:

1. Platforms classify music through mood and function.
2. Generative apps turn those labels into runtime controls.
3. Model-based systems can synthesize the material as well.

This makes functional audio a plausible site for early AI adoption because listeners already accept semantic control, continuous variation, weak work boundaries and limited artist visibility. It remains a possibility rather than an inevitable future.

Questions worth carrying into the revised theory section

• Who defines what Focus, Sleep or Relax sounds like, and which bodies or listening habits does that definition assume?
• Does sensor-driven responsiveness make listening interactive, or does it turn embodiment into an input the service can measure and process?
• Where does authorship sit when source material, rules, runtime realization and later updates occur at different times?
• What counts as musical skill when success may mean remaining unobtrusive, avoiding fatigue and meeting a functional target?
• Does AI change the structure of authorship, or mainly increase the scale, opacity and speed of an existing service model?

Stronger second-half sequence

1. The proprietary service as musical medium: connect interface, production system, runtime and governance.
2. Function labels as musical instructions: show how a semantic category becomes a production goal and runtime control.
3. Embodiment through sensor data: separate bodily influence from deliberate performance or authorship.
4. Musicianship as translation and system design: connect material, rules, evaluation and maintenance.
5. Variable form under centralized control: answer who contributes, who authors and who can alter the field.

The AI outlook then follows without a conceptual jump: model generation changes what produces the material, while the function-first service already defines the relation among listener aim, musical system and platform control.

Sonic consequence rule

Each theoretical claim should name an audible consequence when the evidence supports one. Affordances should lead to duration, density, attacks, transitions, timbre or listener attention. Embodied input should lead to the parameters it may change. Functional value should identify what counts as sonic success or failure. The distinction between rules and models should identify recombination of a bounded palette versus synthesis of new material.

Ownership, licensing, access and governance do not always have an immediate sonic correlate. Do not invent one. Their musical relevance lies in who can change, maintain or withdraw the system that produces the sound.

Current research added on 21 July 2026

• Hesmondhalgh et al. support the interface claim: mood and function frame possible uses without replacing every aesthetic relation.
• Campos Valverde and Lupinacci make the wellness question explicit and discuss Endel’s scientific framing.
• Live Music Models provide a current technical comparison for actual real-time model synthesis.
• Bown remains the main case source because he treats a generative music engine as audio assets, runtime software, interface and organizational work, then asks how that system reshapes artist visibility and service control.

Professor feedback

Guilherme approved the topic, question and links to the course. His feedback sets three priorities:

1. Keep the main question in control of the whole presentation.
2. Place the current case within trends from recent decades without turning the talk into a history.
3. End with current movements in how people make and interact with music.

The deck uses a brief three-part context, a sustained Endel case, a check on claims about passive listeners and a final comparison of rule-based steering with real-time model output.

Evidence boundaries

• Functional listening predates generative apps. Ambient and open-form work, continuous online streams and platform function categories provide context rather than one clean causal history.
• Functional listening does not make music empty or listeners passive. Hesmondhalgh and Campos Valverde show that function, pleasure, memory, aesthetics and attention can coexist.
• Generative, adaptive, personalized, bioresponsive and AI-based describe different mechanisms. The deck does not use them as synonyms.
• Endel is best treated as a proprietary music service with some platform dimensions, not as a complete platform in every sense.
• Endel’s public material supports a bounded runtime process but does not disclose its current node graph, exact mappings, model roles or complete artist workflow.

Claims to avoid

• The Grimes example supports a narrow claim: she supplied recognizable material and revised results while Endel’s team integrated it into the runtime. It does not prove that every artist authors rules.
• A fixed audio or video excerpt can show timbre, density, continuity and intended form. It cannot prove live adaptation, a unique session, sleep effects or causal efficacy.
• Patents show protected possibilities, not shipped code. They stay in backup.
• The Endel focus study measured a model-derived EEG focus score. It did not measure productivity or isolate personalization, and the reported pairwise time-series analysis found no window in which Endel significantly outperformed Apple Music.
• AI is one possible extra layer of runtime variation. It is not the necessary next step and not the whole shift.

Twenty-minute structure

The planned speech totals 17:10. Roughly forty seconds for slide changes and the two excerpts brings it to about 17:50, leaving 2:10 before the twenty-minute limit. Slides 19–24 are hidden, untimed backup slides.

Main-deck notes

Slide 1: Composing the Conditions

Open with the field: generative apps built for focus, sleep and relaxation raise a question about what a musician finishes when the heard sequence remains open.

Slide 2: In Endel, listeners choose a function before a work

Contrast artist, album and track with Focus, Sleep and Relax. The changed first action makes a desired use part of the interface and production brief while leaving room for aesthetic listening.

Slide 3: Three earlier shifts converge in current apps

Use three points of context: ambient and open environmental form, continuous online streams and platform categories such as focus. They converge in current apps without forming one straight genealogy.

Slide 4: Function, generation and adaptation combine in different ways

Endel offers responsive function, (Not Boring) Vibes composes from fragments around daily activity, LifeScore adapts recorded material and Apple offers fixed function playlists. Treat them as a field comparison across several mechanisms.

Slide 5: Research question

State the question once. The rest of the talk asks what becomes variable, what creators make, how listeners still matter and who can alter the system.

Slide 6: Automation moves choices from composition into runtime

Affordances invite and frame actions without determining them. Automation stores ranges, transitions and limits so that the system can make some musical decisions later. A function button frames a task without dictating one listening experience.

Slide 7: The app is one layer of a proprietary music service

Trace material → system → interface → service. Musical mediation distributes the work across assets, code, people and realizations (Born, 2005). Service governance adds subscriptions, data, integrations, updates, policies and artist relations. The unit of analysis is larger than the algorithm.

Slide 8: A responsive Endel soundscape begins with a chosen function

The safe public claim is:

chosen function + permitted context
-> designed musical constraints and mappings
-> one realization

The slide uses Energy because the screenshot and claim refer to the same normal mode. Public sources do not disclose the current mappings or model roles.

Slide 9: Endel presents sleep as a sequence of musical phases

Play 00:40–01:05 from the official Endel Sleep explainer. Listen for continuity through soft attacks, few transients, a stable level and slow timbral change, then notice the weak track boundaries. Keep the company’s description separate from the audible excerpt: the video labels activation, onset and deep sleep, presents its final segment as a continuation through the night and elsewhere describes normal soundscapes as endless. The clip does not prove adaptation or a sleep effect.

Slide 10: The production network designs a bounded possibility space

Artists and the sound team shape voice, timbre and harmony while the sound team and developers define the constraints, mappings and code. The listener selects a mode and supplies permitted conditions, from which the runtime produces one realization. No single actor fixes the heard sequence.

Slide 11: Behavioral listening becomes central to service production

Render many sessions, find fatigue and failure, then revise the limits. Long duration, repetition, edge cases, unwanted attention and incoherent transitions become musical production problems. Behavioral listening joins composition, curation, testing and maintenance.

Slide 12: Endel built Grimes’s material into the runtime

Grimes supplied original vocals, music and stems, then revised the material after auditioning results, as reported in Vogue’s 9 November 2020 interview. Endel’s team built that material into the soundscape. Play 00:20–00:45 from the official AI Lullaby video. The fixed clip shows sonic identity rather than live adaptation, and public sources do not disclose the internal AI process.

Slide 13: The button hides a production network

Recover five kinds of work:

• artist and sound material
• sound design and continuity
• development and rendering
• research, product and interface
• listener choice and context

The slide is an analytic role map rather than a disclosed project task log.

Slide 14: Variable form does not mean equal control

Artists, production teams, software, listeners and context can shape one realization. The service owner and technical team set the field through functions, success criteria, runtime updates, access and the data relation. Generativity distributes causal input, although proprietary ownership still concentrates governance. The question is: who defines what “focus” or “sleep” should sound like?

Slide 15: A listener can seek function and still hear music

Function includes focus, sleep and self-regulation. Aesthetic relation includes taste, memory, pleasure and attention, which can coexist with that function without being exhausted by it.

Slide 16: Function-first interfaces extend beyond generative apps

Compare the same first action across Endel and Apple. Endel turns the goal into a runtime system, while Apple routes it into a curated stream. The interface shift is wider than generation, although the musical mechanism still matters.

Slide 17: Rule-based and AI systems both expand steering

(Not Boring) Vibes shows a second function-first system that composes in real time from musical fragments and responds to daily rhythm, movement, Energy and Presence controls. Lyria RealTime demonstrates model output. Rules already move form to runtime, and AI can add another source of material or variation. Both belong to the current move toward more provisional and steerable form, which does not imply one inevitable AI future.

Slide 18: These services maintain a space of possible realizations

Variable works, scores and open form are not new. The change sits at service level, where a production network distributes tasks while a proprietary service can execute, update or withdraw the conditions after release. This does not turn every musician into a programmer, and generativity does not cause concentrated governance.

Answer the question with four roles:

• Creators: materials and sometimes constraints.
• Production team: rules, criteria and maintenance.
• Runtime and listener: one heard result.
• Service owner: may update or withdraw the field.

Discussion:

1. Where should authorship sit: materials, rules, runtime, or heard result?
2. Who should define what “focus” or “sleep” sounds like?

Backup slides

The backup evidence rests on generative-functional-music-apps, endel-sound-generation, literature/haruvi-endel-focus-study, adaptive-generative-and-functional-music and endel. Full product sources, patent links, videos, image credits and academic citations appear in the speaker notes.

Visible citation audit

Visible footnotes stay limited to claims whose force depends on a historical record, a theoretical concept or an empirical finding. Product pages, app listings, videos, screenshots, patents and general course provenance stay in hidden source notes.

Historical evidence

• Cardinell supplies the period definition of functional music.
• Jones and Schumacher document Muzak’s stimulus ratings and planned rise around worker fatigue.
• The archived ChilledCow watch page dates and describes the first continuous study stream.
• Karakayalı and Alpertan connect function-led listening categories to streaming platforms and user choice.

Theory and empirical evidence

• Magnusson supports the move from a fixed sequence toward a score that remains active during performance.
• Bown treats a generative music engine as musical material, software and production practice rather than an algorithm alone.
• Born supplies relayed creativity and the need to distinguish distributed influence from power over the system.
• Hesmondhalgh and Campos Valverde show that functional listening can retain taste, attention and emotional weight.

Hidden provenance

• Endel’s inputs, modes and production claims remain tied to product material and reporting in hidden notes.
• Videos and screenshots remain evidence for what the presentation shows, not academic authority.
• Patents remain technical context because they describe protected possibilities rather than confirmed current implementation.
• Course slide references remain preparation notes unless the visible slide quotes or directly depends on them.

references