Press Clipping
The “celestial jukebox" era is dead. What's next for music streaming?

The “celestial jukebox" era is dead. What's next for music streaming?
Hey everyone!

As promised, here's the second longform essay of the day, which takes a critical look at the historic "celestial jukebox" metaphor on which much of the modern music-streaming sector has formulated its identity.

It's looong, so I encourage you to find some time to sit with it properly if you can. But tl;dr — I think that the "celestial jukebox" metaphor is irrelevant for how we actually consume music on streaming services today, and offer three alternative metaphors for music consumption and curation that I think will define much of the next five years.

If you're coming from the Water & Music newsletter, welcome!! This is the first Patreon post I've made that is completely free and open to the public, since I wanted to share the ideas below with a wider audience. I would love to hear any specific feedback or questions you have about this piece — simply reply to the email in your inbox, and it'll go straight to me!



The “Celestial Jukebox” Era Is Dead. What’s Next For Music Streaming?

Research references:

The jukebox manifesto (Janelle Brown, Salon, November 2000)
Paradox of celestial jukebox: Resurgence of market control (Hyojung Sun, Creative Industries Journal, January 2019)
Discoverability: Toward a Definition of Content Discovery Through Platforms (Fenwick McKelvey and Robert Hunt, Social Media + Society, January 2019)
What's Next for Marketplace Startups? (Li Jin and Andrew Chen, Andreessen Horowitz blog, November 2018)
A few days ago, I noticed something extremely petty, but nonetheless important to our discussion today: Apple Music changed the number of songs advertised on its homepage from “50 million” to “60 million.”

As you can see in the above screenshots, the homepage frames music streaming as a seemingly infinite, fantastical forest in which users can “lose themselves” at their own whims. Importantly, this rhetoric isn’t new; it’s reminiscent of the decades-old concept of the “celestial jukebox,” which was first imagined in the 1990s as a jukebox that literally lived in the sky (or, in modern parlance, in the cloud?) and could give consumers access to literally any music or media they wanted on their personal devices on demand.

Apple Music first posted about the fact that it had 50 million songs in February 2019 — implying that it added 10 million more tracks to its catalog in a matter of months. And it’s not just Apple who wants to boast its size: Today’s music streaming platforms can’t stop talking about how many millions of tracks they have on their respective apps.

Tidal apparently beat Apple to the 60-million-songs mark in December 2018. SoundCloud boasted in February 2019 that the size of its catalog officially surpassed 200 million tracks. Not only does Spotify now tout more than 50 million tracks as of April 2019, but the company’s CEO Daniel Ek recently confirmed that nearly 40,000 new tracks are now being uploaded to the service on a daily basis — nearly double the previously reported daily volume of 20,000 tracks just over a year ago.

What I want to talk about today is why this collective emphasis on volume is actually irrelevant to how streaming consumption and curation actually work today — and why we need a new metaphor for understanding music streaming in 2020, instead of aggressively clinging to the “celestial jukebox” with which we once fell in love.

I’ll use myself as one example. According to my Spotify Wrapped, I’ve listened to around 19,000 minutes of music so far in 2019. Assuming that the average song lasts 3.5 minutes long, I could have listened to a maximum of 5,500 songs this year — or 0.01% of Spotify’s entire catalog of 50+ million tracks.

While 5,500 is a pretty big number, in the context of Spotify’s catalog, that’s far from “celestial.”

In fact, when I stream music, I don’t really feel like I’m wandering through an endless forest at all. Instead, I end up making my listening decisions, either consciously or subconsciously, in an environment that has several external commercial and technological forces whittling the infinite forest of music streaming down to only a select few trees for me to choose from.

In short, in the modern music business, the “celestial jukebox” that captured our imagination in the early 2000s has since been pinned down to earth and winnowed by a combination of market control and technological control.

The concept of the “celestial jukebox” goes back at least to 1994 (five years before Napster was founded), when Stanford Law School Professor Paul Goldstein outlined the concept in his book Copyright’s Highway.

Goldstein’s description of a celestial jukebox was literally astronomical: “a technology-packed satellite orbiting thousands of miles above the earth” that could “give tens of millions of people access to a vast range of films, sound recordings, and printed material, awaiting only a subscriber’s electronic command for it to pop up on his television or computer screen.”

A device giving you access to a vast range of content at your command? That sounds a lot like the first MP3 players. In fact, one year before Apple released its inaugural iPod to the public, Salon suggested that a “celestial jukebox” was not an otherworldly, imaginary, floating object, but rather a real hardware device that could thrive in the market at the price point of around $150 upfront, plus “$10-$20 a month as a flat subscription fee for access to the music … You'd never have to buy a CD [and] the ease-of-use would make the price painless.”

The following 20 years saw various new music devices emerge, each of which commentators would peg as a more definitive “celestial jukebox” than the previous ones. Writing for WIRED, Eliot Van Buskirk gave the “celestial jukebox” nickname to the Verizon V Cast and Sprint Music Store (with around 1 million and 400,000 songs at the time, respectively) in 2006, then to the Omnifone MusicStation (with around 1.5 million songs) in 2008. Van Buskirk touted the benefits of moving from “impulse buying,” which could quickly make an unintended dent in people’s wallets, to what he called “impulse listening,” which was made more affordable via a WiFi-enabled subscription service for a monthly fee.

But none of these devices really got off the ground in terms of long-term sales. It wasn’t until Spotify’s early years — and subsequent swift rise — that onlookers truly recognized the mainstream acceptance of the “celestial-jukebox” model as the standard for music consumption, rather than as just a fringe fantasy. Indeed, in light of paid streaming subscriptions driving much of the recorded-music industry’s growth over the last five years, it’s normal to assume that the modern music business thrives off of giving consumers essentially infinite choice over what to listen to.

Reality, of course, is not so simple.

Streaming services will jump immediately at the opportunity to advertise the millions of tracks available on their platforms; they’re much more hesitant to promote the tyranny of choice, and their subsequent control over user behavior, that results. Popularized by psychology professor Barry Schwartz, the “tyranny of choice” refers to how human beings seems to grow less happy and less satisfied the more they are granted a sense of choice and personal autonomy.

A practical, non-musical example: If you’ve eaten out at a restaurant, you’ve probably encountered a menu that seemed to have just too many items on it. You try to scan the menu as quickly as you can, but — quickly overwhelmed by the amount of choice — you slip away from logical or adventurous decision-making, and instead lean towards the familiar, ordering a dish you’ve already eaten dozens of times elsewhere.

The “celestial jukebox” of early streaming, while idyllic in theory, presents a similar conundrum: the more choices we face for what to listen to or watch as users of these platforms, the more paralyzed we become and the more we tend to flock towards what we already know. Or, the less we remember the vast swath of music we end up consuming. As Motive Unknown’s Darren Hemmings put it: “By giving us everything of everything, we overload and take nothing of anything, overwhelmed in the face of it all.”

Over the past decade or so, the music industry has tried to correct for this choice paralysis by implementing both market controls for filtering which artists gets funded and promoted to streaming services, and technological controls for filtering how those artists’ works get delivered to listeners.

I describe each of these types of controls below:

In a world of chronic abundance, gatekeepers earn more power, not less. The gatekeepers of today’s music industry might not be wearing the same clothes as they did 10 to 20 years ago, but they’re still here.

Since the mid-20th century, the recorded-music business has been rife with gatekeepers. From A&R scouts and agents, to record promoters, to radio stations, newspapers and other media outlets, to, of course, the record labels themselves — music gatekeepers trusted each other to reduce their own, collective risk, while preserving their power by preventing potential competition from entering the market.

The rise of streaming has led many to believe that major labels’ control over distribution has loosened, making music discovery more organic and democratized. This is somewhat true if you look at the numbers: According to MIDiA Research, unsigned artists comprise the fastest-growing segment of the music business, outpacing both major and independent labels.

But again, in an age of chronic abundance, listeners crave guidance on what to listen to — only further underscoring the importance of tastemakers who influence audiences via blogs, playlists, radio shows, YouTube channels and other forms of media.

Moreover, the major labels haven’t gone anywhere. Each of them now owns a handful of playlisting and distribution companies. Two of them still own equity stakes in Spotify. All three of them have licensing deals with Spotify that guarantee a certain amount of visibility and market share on Spotify’s editorial playlists.

Researcher Hyojung Sun mapped out this new landscape of music gatekeepers in her paper “Paradox of celestial jukebox: Resurgence of market control”:

[Image source]

The above diagram is slightly outdated, in that third-party playlisting companies are no longer as influential in driving mainstream tastes as they used to be. This is in part because virtually every streaming service now prioritizes their own-and-operated playlist properties over third-party ones on their respective user interfaces, which is another form of market control (more on that later).

Nonetheless, in the current streaming paradigm, major labels still use their catalog sizes and financial prowess as leverage to preserve their own power by staking out both “shelf space” on streaming services’ playlists and “virtual space” in listeners’ heads. As Sun writes: “The limitation imposed by physical space in the conventional setting is replaced by the limitation of virtual space — attention.”

Under this new form of market control, Sun continues, digital services simply turn into “a ‘tollbooth’ through which copyright meticulously monetizes every instance of music listening … but does not, in turn, foster innovation.”

This is a key revelation: while technological advancements and trends come and go, market control has largely stayed the same, and even further consolidated, in music over the past ten years. And streaming services like Spotify and Apple Music need to align with these incumbent structures in order to stay afloat, in part because of their licensing deals, limiting their ability to innovate over time.

In addition, as consumer-facing tech platforms, streaming services also want to control their own audience relationships to ensure they can achieve their own internal goals (around user retention, engagement, etc.), independently of any market control that the music industry demands. This leads us to the next section...

Zooming out a bit, music curation actually fits into a wider history of tech platforms and marketplaces shifting from a hands-off to an explicitly hands-on approach to curating products and services for sale. In this vein, the music industry has a lot to learn from the tech industry.

Back in the 1990s, online marketplaces — sites on which everyday people can buy, sell and exchange stuff — operated on the premise of providing a seemingly infinite, unmanaged catalogs of products, services and destinations, on a platform that connects buyers to sellers.

The poster children of this era are Craigslist and Yelp, which were popular at their outset because volume alone was the perceived source of value. They just listed so much more stuff than any other sites — the equivalent of a “celestial Yellow Pages.”

But along with that volume-based approach came a crucial obstacle for buyers and sellers: they were the ones who had to vet each other’s quality, with the platforms largely freeing themselves of liability in the case of a low-quality transaction. “It used to be the case that marketplaces could add a ton of value just by aggregating any and all supply and demand,” Li Jin, partner at venture-capital firm Andreessen Horowitz, explained on Twitter. “The onus was on the users to filter using the available information in the marketplace to choose what they wanted or needed.”

Eventually, as specific kinds of products and services that required a lot more trust started to enter these infinite marketplaces (e.g. real estate, childcare, transportation, food) customers wanted platforms to take on more responsibility and do more quality control and curation themselves.

This has led to a dramatic increase over the past decade in more “managed” marketplaces — i.e. in marketplaces that take on more work, not less, to facilitate the delivery of the product or service (e.g. by interviewing and vetting suppliers directly, or automatically setting prices).

Andreessen Horowitz outlined this evolution on a timeline earlier this year, pictured below:

[Image source]

As illustrated above, there are three primary categories of tech marketplaces that emerged after the “celestial Yellow Pages” model:

“Unbundled Craigslist” marketplaces — which broke down the Craiglists and Yelps of the world into vertical-specific offerings, such as for childcare, Angie’s List and TaskRabbit for home services and Fiverr for business services such as graphic design and copywriting.
“Uber for X” marketplaces — which are not only vertical-specific but also add on the important feature of automatically matching supply and demand; think Uber and Lyft for transportation, Doordash and Seamless for food and Glamsquad for on-demand stylists. Not only are transactions more efficient, but the platforms themselves also now take on a more prominent role as the service provider than the person or business that’s actually performing the service. As a result, sellers no longer have the benefit of owning their relationships with buyers, and can hence feel alienated (this conflict is at the heart of the fight for gig workers’ rights around the world).
“Managed marketplaces” — which are most relevant to our discussion today, and take on more responsibility in vetting services that might be too complex and/or require too much trust to warrant simply a one-click, automated user experience. Key examples include Honor for elder care, Trusted for childcare, Wonderschool for in-home daycare and Opendoor for home selling.
Taking all of these categories into account, one realizes that Spotify has gone through a similar transition in its history — from a “celestial jukebox” offering, to a much more tightly managed and controlled marketplace for music curation.

When Spotify first launched in the U.S. in 2011, it was advertised simply as an all-you-can-eat, on-demand listing of music, with almost no additional benefits around platform-driven curation, according to press coverage at the time. It wasn't until outside services like Songza and Tunigo came in with more superior curation options that Spotify recognized the urgent need to step up its own curation game. (Spotify ultimately acquired Tunigo, which significantly altered its approach to curation; the book Spotify Teardown provides a good deep-dive into this transition.)

Today, Spotify is the combination of what investors would call a “Managed Marketplace” and “Uber for X” — i.e. partially automating the match of supply (e.g. artists’ songs) and demand (e.g. listeners’ tastes) through algorithmic recommendations, while also offering hundreds of editorial playlists that frame the platform itself as the primary facilitator of curation, rather than artists or users.

If you are a Premium subscriber to Spotify, you are not paying for music. You are paying for access to Spotify’s service offering, which positions music as a more utilitarian form of entertainment that can squeeze itself into any moment in your day, whether you’re working out, traveling, resting or eating. As I wrote previously, this creates a new type of superfan that prioritizes the streaming service over the artist — just as “managed marketplaces” in tech prioritize the platform over the seller as the actual service provider.

Interestingly, taking on this responsibility as a streaming service seems to be the new table stakes for staying competitive at all. Apple Music, Tidal, Deezer and SoundCloud all launched their own versions of personalized and editorial playlists in Spotify’s wake (e.g. SoundCloud Weekly, Deezer Flow, Apple Music’s My New Music Mix).

In general, music platforms today are underrated in their power to drive discoverability — which Fenwick McKelvey and Robert Hunt defined in a paper as “how content discovery platforms coordinate users, content creators, and software to make content more or less engaging.”

Music streaming companies have enormous influence over which artists get discovered, and in turn how our tastes as music fans evolve. According to YouTube’s own data, algorithmically recommended videos account for 70% of viewing time on their site. In 2016, it was revealed that there were several thousand artists on Spotify for whom Discover Weekly accounted for over 50% of their total streams. And as the story about Rory Fresco and Kanye West demonstrated, algorithms can unwittingly help break an artist into the mainstream.

But at the end of the day, platforms own these experiences and will optimize them for particular outcomes that favor their bottom lines and maximize user engagement. This leads to a constant mindset among several independent artists of “chasing the algorithm” to rack up streams, views and, hopefully, fans — and, in the process, being increasingly reliant on a system over which they ultimately have little to no control.

So just to reiterate: The decentralized, all-you-can-eat “celestial jukebox” concept has been significantly reduced, through a combination of market- and tech-driven controls, to a much smaller, centralized selection of songs for consumers to browse from day to day. Hence, we need different, more appropriate metaphors for describing what streaming has really become.

In general, when thinking about the future of music streaming and curation, I think there will be a fundamental recalibration in where we source the value of curation.

The appeal of the “celestial jukebox” framework was the straightforwardness of objective, measurable value. Platforms working under this framework attempted to gain competitive advantage over each other through concrete measures such as growing their song catalogs, nailing exclusive content and improving audio quality. It’s easy to debate whether music is “good,” but slightly more difficult to argue with something as quantifiable as “60 million tracks" or "dozens of exclusive videos."

I would argue that because the novelty and innovation around that framework has largely worn off, future music-driven content companies will increasingly compete on the vector of subjective, perceived value.

The competitive advantage of one platform versus the other in this second framework is highly dependent on people, rather than on algorithms or other tech-centric factors. Individual people have preferences, tastes and emotions that are complex and difficult to reduce to a quantifiable data point. Value in this environment becomes highly relational, contextual and situational.

It also becomes much more conducive to niche social networks — a phenomenon that even the world’s largest social-media platforms are doubling down on (e.g. Facebook investing more in groups and region- and interest-specific community development).

It’s quite the opposite of celestial and otherworldly; it’s down-to-earth, and human.

There are three potential paths forward under this umbrella of more subjective, perceived value: artist-centric, listener-centric and vertical-centric. Please bear with my stick-figure visualizations.

While music streaming brought the aggregate recorded-music industry back to growth, it’s also led to what some people in the music industry have called the “dry streams paradox” — whereby artists can rack up significant volumes of streams by way of inclusion in a large playlist or algorithmically-curated radio station, with little actual interest or recognition of the artist from listeners.

Artist-centric curation models try to break down this unexpected barrier between artists and fans, by putting the artists themselves at the core of curation, and making the curation of media and experiences truly “direct-to-fan,” instead of relying purely on hands-off algorithms to do the job.

There are already several forms of artist-centric curation taking place in the music industry that will only continue to grow in the future:

Artist-run collectives and labels — e.g. ODESZA’s Foreign Family Collective, J. Cole’s Dreamville.
Artist-curated festivals — e.g. Tyler the Creator’s Camp Flog Gnaw, Florida Georgia Line’s FGL Fest.
Artist-hosted podcasts and radio shows — e.g. Gimme Radio, Song Exploder, Blonded Radio.
Artist-written newsletters — e.g. George Ezra’s Journal, Nick Cave’s The Red Hand Files.
I refer to artist-centric curation as a “tribe” model because the goal is not to maximize scale, but rather to foster and strengthen ties with a small, highly loyal community of fans and followers.

This isn’t to say that incumbent, mass-market streaming platforms don’t already have features that cater somewhat towards an artist-centric curation model. For instance, Spotify maintains a sprawling series of “This Is” playlists centered around individual artists’ catalogs (e.g. “This Is James Blake”) — although the way in which they are structured is somewhat opaque and, in most cases, beyond the control of the actual artists themselves.

In spring 2019, Spotify also introduced a new way for artists to share editorial playlists on Spotify, such that “anyone who clicks a unique link shared by an artist will see a personalized version of the playlist with that artist’s track as the first song.” The messaging around this feature is not only that it gives more control to artists, but that it also has a tangible positive impact on their streams.

“These personalized editorial playlists increase the number of artists featured on playlists by 30% and the number of songs listeners are discovering by 35%,” wrote Spotify, who continued: “After discovering a song through a personalized editorial playlist, the number of listeners who then seek out the track on their own for repeat listens is up by 80%. In fact, the average number of times a listener saves a track is up 66% — all of which is good news for artists.”

[Image source]

In the artist-centric case, however, it’s arguably not enough just to change the surface-level curation interface: you also need to change the underlying economics of how artists are compensated for their work.

As I’ve written recently, for an incumbent, mass-market streaming service to become truly “artist-centric,” it needs to shift its financial structure from monetizing aggregation of content, to monetizing on the level of individuals and their direct relationships with their audiences.

Darren Hemmings has suggested that the music industry develop “an equivalent to the Fair Trade mark in coffee,” which would indicate that “when you spend money on an artist, they are the main beneficiary and not the platform.” Some services like Deezer will also be piloting a “user-centric licensing” model in 2020 that will compensate artists for every individual time a user streams their music, instead of paying them less often based on a pro rata model.

Some people in the music industry would argue that the dominant music-streaming paradigm is already listener-centric, because of its emphasis on portability and personalization.

But this form of personalization is highly individualistic and anti-social. There’s no way on these mass-market streaming services for fans to interact genuinely with each other, let alone with their favorite artists. Those looking to foster that deeper connection are practically forced to look elsewhere.

A few new musical experiences are emerging that try to create a more explicitly social, networked layer over the isolated listening experience that currently exists, in a way that rewards sharing and community-building.

This is a particular area of interest in the world of music and blockchain, in which companies like Choon and Lightstreams are working on solutions for decentralizing music curation and compensating users financially for engagement around their curation, regardless of their popularity or social stature. The app Stationhead is trying to recreate the “pirate radio” experience by allowing users to broadcast their voices live over a music stream to their friends.

And TikTok — easily one of the apps of the year, with a major profile in almost every major publication — is renowned for its lattice-like, user-driven, algorithmic and relatively decentralized approach to content curation, regularly creating semi-viral moments for the most unsuspecting artists.

In general, I’ve designated the models above as “listener-centric” because they tie curation much more to an individual user’s transparent social graph, rather than to crowdsourced consumption behavior that’s analyzed in a black box.

Another emerging, and somewhat controversial, method of listener-centric music curation is literally tying curation to users’ physiologies, by incorporating biometrics into algorithmic recommendations. For example, consumer-facing startups like Weav and Endel are working on experiences that can change a song’s tempo, instrumentation, volume and other factors based on users’ moods, heart rates and immediate activities (e.g. sleeping, working, running).

Interestingly, some tech writers in the recent past have called for more consumer control over the saturated abundance of content afforded by tech, by essentially eliminating tech altogether.

For instance, The New York Times’ music-industry reporter Ben Sisario shared in May 2019 that his preferred way to listen to music was on CDs, because the format had “no ads, no privacy terrors, [and] no algorithms.” Similarly, in March 2019, Damon Beres, Deputy Editor of Medium’s OneZero publication, called for future tech companies to draw inspiration from the classic iPod clickwheel in their future UX designs — i.e. downsizing functionality for the sake of simplicity. The “everywhere-ness [of tech] has become a burden to the modern internet user … What may be more useful, now, at the turn of a new decade, is a renewed focus on individual devices that excel at individual purposes,” wrote Beres.

One wonders, though, whether this retrograde approach will actually be ideal for music culture at large. An especially powerful listener-centric model for the future would combine this highly personal, subjective, downsized approach to taste that Sisario and Beres espouse, with a smooth and decentralized lattice model for curation online that keeps fans at the core.

A vertical- and activity-centric model for music curation would be analogous to the “Uber for X” marketplaces that emerged in the early 2010s: focusing on a specific vertical or physical sphere of music listening, while having the platform play a more proactive and authoritative role in content curation.

By “vertical-specific," I’m not thinking so much of genre-specific platforms, like Audiomack for hip-hop or Gimme Radio for metal. Rather, I’m thinking of platforms that own a specific cluster of interests and/or day-to-day tasks (e.g. fitness, gaming, film, fashion, food), as they relate to music.

Importantly, this new model of curation isn’t purely “functional” — otherwise, we’d just be recreating many of the woes of mass-market streaming services that brought us here in the first place (hello, “fake artists”?). Instead, it tries to combine the functional focus on interests and tasks, with more higher-level cultural expertise tied to the current zeitgeist (what some investors have deemed “product-zeitgeist fit”).

For instance, meditation startup Calm is far from just an algorithmically-driven app for musical relaxation; its music team is already working with major artists on releasing remixes and exclusive content for its app. Five Vectors is taking a similar hands-on approach to building music-curation infrastructure catered specifically to gamers.

One important question that emerges from these artist-centric, listener-centric and activity-centric curation models is whether they are actually truly disruptive or transformative, in relation to the current streaming paradigm.

All of these alternative metaphors for music curation rely on establishing some sort of context — be that a social context, such as a lattice of relationships, or a personal one, such as a daily routine. But in our modern era, human context is nearly impossible to decouple from technology.

From mobile phones (which over five billion people in the world now own), to computers and laptops (which are selling around 58 million units globally per quarter), to smart speakers (which sold 28.5 million units globally in Q3 2019, 36% of which came from Amazon) — our everyday contexts, encompassing the way we do work, communicate with each other and discover new information and media, are largely in the hands of tech companies, and especially hardware companies.

Whoever owns these contexts, as determined by the scope of the technological configurations they own in our daily lives, will arguably rule the future of tech, business and media.

In this vein, Spotify is arguably far behind Apple, Amazon and Alphabet as far as ownership is concerned; there’s a reason Spotify is giving out so many Google Home Minis to subscribers for free.

Herein is the paradox: If the fight for the future of curation is about the fight for the best understanding of subjective value, it’ll also inevitably be a much more product-driven fight for who can own the biggest piece of our daily lives through technology.

Just as gatekeepers often become more powerful, not less, in a world of abundant content, technology companies will become more powerful, not less, in a world of abundant contexts. 🌊

Thanks to Yash Bagal for his research assistance on this piece!