What is musen?
Version: v3.0.0 Last updated: February 2026
1. Reframing audio, not diagnosing a problem
The evolution of digital audio over the past decade has been driven by scale. Catalogs have grown larger, access has become frictionless, and algorithms have become increasingly sophisticated at ranking and recommending content.
Despite this progress, the dominant mental model has remained largely unchanged: audio is treated as a collection of discrete items to be selected, queued, or skipped. Whether expressed as playlists, feeds, or recommendations, the underlying assumption is that better choices lead to better listening.
musen starts from a different premise. Audio listening, especially in everyday life, is rarely about making choices. It is about continuity, context, and flow. People listen while working, commuting, resting, or living their routines. In these moments, audio functions less like a catalog and more like an environment.
Radio historically addressed this by removing the burden of choice and replacing it with programming. What musen proposes is not a fix to a broken medium, but a continuation of this idea using modern computational systems.
The shift is not from bad solutions to good ones, but from selection to experience, and from discrete optimization to continuous orchestration.
2. The core insight: radio is a continuous computation
musen is built on a simple but non-trivial insight: Radio is not a sequence of choices. It is a computation unfolding over time.
Accordingly, musen does not ask what song should play next. It asks: what should be happening in this radio stream right now
Answering this question requires reasoning over:
current context (time of day, session conditions)
recent history (what has already played)
long-term listening patterns
editorial structure (energy, pacing, speech density)
This cannot be solved by recommender systems alone, nor by generative models in isolation. It requires an orchestration system that treats time, continuity, and adaptation as first-class concerns.
musen refers to this class of systems as AI Radio.
3. How musen computes radio
At runtime, musen operates as a closed intelligence loop:
This loop runs automatically while a radio is active. The listener does not need to manage it. Adaptation emerges implicitly from listening itself.
Two editorial agents operate within this loop:
AI DJ — curates fixed segments for on-demand radio listening
AI Host — curates live radio streams in real time
These agents receive candidate selections and flow directives, then make the final editorial choices that define the listening experience. Intelligence guides the process, but editorial judgment remains explicit.
4. Long horizon memory as infrastructure
Continuous radio cannot adapt meaningfully without memory.
Short-term signals such as skips or likes are insufficient to understand when a user listens, how long attention is sustained, tolerance for speech versus music, and preferred energy patterns over time.
musen therefore treats long horizon memory as infrastructure, not as a feature.
Memory is session-based rather than click-based, time-aware and context-sensitive, intentionally decayed rather than permanently accumulated.
This allows the system to adapt gradually without locking users into static profiles. Over time, the radio learns how to sound right for a listener, without requiring constant interaction.
This is the technical foundation of musen’s “just press play” experience.
5. Economics that follow listening
Most streaming platforms pool subscription revenue and redistribute it through opaque, popularity-weighted mechanisms. Value does not reliably track attention, and creators are incentivized to optimize for algorithmic placement rather than listener experience.
musen adopts a different model.
Economic value is generated continuously as listening occurs and is allocated based on listening time, not on plays or clicks. Each listener funds only the radios they actually listen to. There are no global pools and no cross-subsidization.
Value is divided along established radio lines:
Music rights holders are compensated based on the duration their tracks are played
Radio creators and curators are compensated for editorial and broadcast performance
This structure is resistant to streaming manipulation, aligns incentives around sustained attention, and allows creators to earn without algorithmic gaming.
6. Why musen, why now
musen is not a feature that can be added to an existing streaming platform.
It requires treating radio as a continuous system rather than a catalog, embedding long horizon memory at the system level, adopting time-native economic accounting, and enforcing conservative legal boundaries around content and AI usage.
These choices create compounding advantages. As listening history accumulates, radios improve. As radios improve, switching costs become experiential rather than contractual. As regulation around AI and media tightens, systems designed for transparency and auditability gain an advantage.
musen is designed to become more defensible over time, not less.
7. Closing perspective
musen is not trying to optimize music selection.
It is building an AI Radio system where listening is effortless, adaptation happens over time, value follows attention, and intelligence serves experience rather than engagement metrics. If AI meaningfully reshapes audio consumption, radio will not be solved by better recommendations alone. It will be solved by systems designed to compute experience continuously. musen is built for that future.
Last updated