# Why Spotify was called an innovative startup

> 2026-11-08 · Essays · engineering culture
> https://hyna.dev/en/posts/spotify-startup-spirit

The Netflix drama [The Playlist](https://en.wikipedia.org/wiki/The_Playlist_(TV_series)) tells the story of Spotify's early years in six episodes. Unusually, the narrator changes every episode. And as a [MovieZine interview](https://www.moviezine.se/nyheter/vad-ar-sant-i-the-playlist-gizem-erdogan-ulf-stenberg) points out, at the end of each episode one character looks into the camera and says that isn't quite how it happened, then brings out their own version in the next one. Each episode pushes back against the one before.

The narrator of the last episode, episode 6, is Bobbi, a fictional artist. Its ending pulls back to show the blue screen on set, revealing that what you just watched never happened. Netflix itself called the show a *fictional drama series* in its [announcement](https://about.netflix.com/en/news/the-playlist-launches-on-netflix-on-october-13-watch-the-teaser-here).

![The episodes of the drama The Playlist laid out in six rows like a playlist. Each row has a number, the episode title and that episode's narrator: 1 The Vision, founder Daniel Ek; 2 The Industry, music industry exec Per Sundin; 3 The Law, lawyer Petra Hansson; 4 The Coder, coder Andreas Ehn; 5 The Partner, co-founder Martin Lorentzon; 6 The Artist, artist Bobbi. To the right of the list, five teal arrows curve up from each row to the row above, each labelled 'next episode rebuts'. Only the row for episode 6 carries a teal 'fictional character' tag, and a note at the bottom says the Netflix press release called the show a fictional drama series.](/images/posts/spotify-startup-spirit/six-voices-en.svg)

*The narrators of The Playlist's six episodes. Each episode opens by overturning the story of the one before.*

As drama, it is a clever structure. But if you want to know why Spotify was called an innovative startup, the show can't serve as evidence, because it never tells you whose version is right. So in this post I won't lean on scenes from the drama. Instead I want to read the records that people at Spotify left themselves, at the time.

Let me put the answer first. My conclusion from reading those records is that Spotify's spirit did not live in the org chart of squads and tribes. The company made what to do, and why, as clear as a number, and left how to do it widely to the teams. In exchange, it built things so that any failure reached only a small area, which meant teams could fail fast without fear. And the people at Spotify saw this way of working as a snapshot of the moment rather than a finished model. That last attitude is the least known part, but as I see it, it is the most important.

## A 200-millisecond goal

At the [Fortune Brainstorm Tech conversation](https://fortune.com/2011/07/21/brainstorm-tech-video-spotifys-daniel-ek-on-transforming-digital-music) in 2011, Ek said that what Spotify obsessed over was the time between thinking of a song and hearing it play. *if we can get people to play it within 200 milliseconds, the human brain actually doesn't even perceive that there's a latency.* The feeling he was after was iTunes with all the world's music on your own hard drive.

A paper from the year before shows this was more than a marketing line. Spotify engineers Gunnar Kreitz and Fredrik Niemelä presented [Spotify – Large Scale, Low Latency, P2P Music-on-Demand Streaming](https://people.cs.umass.edu/~phillipa/CSE390/spotify-p2p10.pdf) at IEEE P2P 2010. The paper measured real playback over one week. The median playback latency was 265 milliseconds and the 90th percentile 1,047 milliseconds.

![Two horizontal bars on an axis running from 0 to 1,100 milliseconds, with a orange dashed line crossing them vertically at 200 milliseconds, labelled 'Target 200ms (Ek, 2011)'. The upper bar is the measurement from the week of 23–29 March 2010. From 0 to the median of 265 it is a solid teal bar labelled 'half of plays'; up to 515 (the 75th percentile) an outlined segment labelled 'next 25%'; up to 1,047 (the 90th percentile) a dashed segment labelled 'next 15%'; the remaining 10% beyond that is marked only in text outside the bar. The lower bar is a different week when next-track prefetching was switched off by a bug, and its 'half of plays' stretches to a median of 390, further past the target line. The thing to notice is that in the upper bar only the solid part ends just past the target line, while the dashed segment runs past one second.](/images/posts/spotify-startup-spirit/latency-target-en.svg)

*The 200-millisecond target and the 2010 measurements. The median was close to the target, but one play in ten took longer than a second.*

The 265 milliseconds includes plays of tracks already stored on your own computer. The paper itself notes that more than half the data in the measurement period came from the cache, so the median most likely measures mainly the time spent on copy protection (DRM) checks and local processing.

What I find interesting is the means used to hold that number. The Spotify desktop app did not get music only from servers. It also ran a P2P network that fetched pieces of a track from other users' computers that had it, and it downloaded the next track before the current one ended. For a few weeks a bug switched off next-track prefetching on every client. In one week of that period the median was 390 milliseconds, higher than the 265 of the measured week. An unintended experiment showed what prefetching was worth.

Then in 2014 Spotify wound down this P2P network itself. In a [TorrentFreak report](https://torrentfreak.com/spotify-starts-shutting-down-its-massive-p2p-network-140416/), a Spotify spokesperson said *We're gradually phasing out the use of our desktop P2P technology*. The report took this to mean that central servers alone would now handle the work P2P had been doing.

I think this moment shows Spotify's first trait best. The goal was set clearly as a number, the time a person feels, and the means were used only for as long as they fit that goal. Even P2P, which had once held the service up, was put down once a better way of meeting the goal came along.

## Set what to do, and the how can be left to the teams

In 2012 the agile coaches Henrik Kniberg and Anders Ivarsson wrote the white paper [Scaling Agile @ Spotify with Tribes, Squads, Chapters & Guilds](https://blog.crisp.se/wp-content/uploads/2012/11/SpotifyScaling.pdf). It is the original source of what people usually call the Spotify model. At the time Spotify had more than 30 teams in three cities.

The paper's first unit is the squad. It is similar to a Scrum team, and it is *designed to feel like a mini-startup*. A squad sits together, has all the skills and tools it needs from design through development and testing to release, and decides for itself whether to use Scrum or Kanban. The ideal was a squad that didn't wait on other squads and talked directly with its stakeholders.

Several squads together make a tribe. Tribes were sized around Dunbar's number, the idea that a person can keep up social relationships with about 100 people. The reasoning was that beyond that size, rules, bureaucracy and politics start to grow. People doing the same kind of work across squads, for example the backend developers within one tribe, were grouped into a chapter. A chapter always stays inside a tribe, while communities of interest that cut across the whole company, like a tester guild, were called guilds.

![Two large boxes side by side are Tribe A and Tribe B, each labelled 'about 100 people (Dunbar's number)'. Each tribe holds three tall squad boxes, and each squad has three circles standing for people, one on each of the Backend, Web and Test rows. In each tribe, the teal circles on the Backend row are wrapped in their own teal horizontal band to form a chapter, and that band ends inside the tribe box. The orange circles on the Test row are wrapped in a single orange dashed band, and this guild band runs across both tribe boxes. The difference to see is that chapters stay inside a tribe while guilds cross tribe boundaries.](/images/posts/spotify-startup-spirit/squad-tribe-chapter-guild-en.svg)

*Squads run vertically, chapters run across, and guilds cross between tribes.*

The passage where the paper explains why chapters and guilds are needed still reads as candid. If squads were fully autonomous and never talked to each other, there would be no reason for the company to exist, and *Spotify might as well be chopped into 30 different small companies*. Chapters and guilds were the glue that held those thirty together as one company.

There is a sentence on the paper's first page that has to be read along with all this. *Disclaimer: We didn't invent this model.* The authors say up front that they did not invent the model, and they describe the paper as *only a snapshot of our current way of working - a journey in progress, not a journey completed*. Then they add one more sentence. *By the time you read this, things have already changed.*

A better guide than the shape of the org chart to what it was trying to do is [Spotify Engineering Culture Part 1](https://engineering.atspotify.com/2014/03/spotify-engineering-culture-part-1), an animated video Kniberg made in 2014. The video introduces squads as *loosely coupled but tightly aligned* and then draws a 2×2 grid.

The horizontal axis is autonomy and the vertical axis is alignment. When both are low, you get a micromanagement culture where people just do as they're told, with no higher purpose. With high alignment and low autonomy, leaders tell teams not only what problem to solve but also how to solve it. With high autonomy and low alignment, each team does whatever it likes and they all run in different directions, and the product, in the video's words, turns into a Frankenstein. Where Spotify wanted to be was the square where both are high: leaders focus on which problem to solve and let the team work out how. The video calls this square *aligned autonomy*.

![A 2×2 grid with autonomy (low to high) on the horizontal axis and alignment (low to high) on the vertical axis. Bottom left is 'Micromanagement: no higher purpose, just do as you're told'. Top left is 'Dictates what and how: the leader names the problem and also decides how to solve it'. Bottom right is 'Teams pull different ways: the leader is helpless, the product is a Frankenstein'. Only the top right square is filled orange and reads 'Aligned autonomy: the leader decides what to solve, the team decides how'. Below the grid is Kniberg's 2014 line, alignment enables autonomy: stronger alignment lets you give more autonomy.](/images/posts/spotify-startup-spirit/aligned-autonomy-en.svg)

*The 2×2 from the 2014 video. What and why come from above; how comes from the team.*

What I think matters most in this grid is the sentence the video adds right after it: *alignment enables autonomy*. Spotify did not put autonomy and alignment on opposite sides of a scale, where gaining one means losing the other. It saw them the other way round: the clearer what you are doing and why, the more widely you can leave how to the team. The 200 milliseconds from earlier is a good example. If the goal is set as a number, the delay a person feels, the team can pick the method, whether that is P2P or prefetching.

But right after the grid, the video also names the price. *one consequence of autonomy is that we have very little standardization.*

## Keeping the reach of failure small

[Spotify Engineering Culture Part 2](https://engineering.atspotify.com/2014/09/spotify-engineering-culture-part-2), released that September, briefly recaps Part 1 and then opens its main topic with a line from Ek. *we aim to make mistakes faster than anyone else.* The video spells this out as fail fast, learn fast, improve fast, and calls Spotify a *fail-friendly environment*, meaning it cares more about recovering from failure than about avoiding it.

But being friendly to failure only works with a condition attached. The video states it like this. *failure must be non-lethal or we don't live to fail again.* That is where the idea of a *limited blast radius* comes from. The architecture is kept decoupled enough that one squad's mistake affects only a small part of the system, and new features go out first to a very small share of users, who are watched before the rollout widens.

The release approach in Part 1 is what keeps that radius small day to day. Each client app has a release train that leaves on a fixed schedule, usually every one or three weeks. If trains leave often and reliably, there is little need to plan far ahead. In the video's words, you just show up at the station and take the next train. If features A, B and C are done and D is still being built, the train leaves carrying all four, and the unfinished D is hidden behind a feature toggle.

I built this relationship as a train station. Each car, A, B, C and D, has a toggle lever on the platform, and when a car's lamp glows amber, that feature is visible to users. At first only the finished A, B and C are on, and D, still in progress, is off. When the clock goes round once, the train leaves with all four cars and waits for nothing. Try pulling D's lever before D is finished. Only D's lamp flickers red and smokes, while the other cars and the departure time stay the same. Push the lever back and the fault disappears at once.

> Interactive demo: Spotify’s release train as a train that leaves on a fixed schedule: all four cars (A, B, C, D) board whether their feature is done or not, and only cars whose platform lever (feature toggle) is on light their lamp and are visible to users. Switching on an unfinished feature faults that car alone, and pushing the lever back clears it at once. — https://hyna.dev/en/posts/spotify-startup-spirit

Spotify also wrote this attitude down as values. The [Band Manifesto](https://web.archive.org/web/20260328031341/https://www.lifeatspotify.com/being-here/the-band-manifesto), which hung for a long time on its careers site Life at Spotify, listed five values. Under Innovative it said *We move fast and take big risks*, adding that every mistake holds a valuable lesson, so there is no need to fear failure.

A culture that encourages failure brings boldness to mind first, but as I see it, what actually did the work was dull machinery like release trains and feature toggles. The train doesn't wait, so one feature can't hold up a whole release, and with toggles, a half-built feature can ride along without users seeing it. Because everyone knows a failure is shut inside D, the team working on D is less afraid too. This machinery is also what let the video say *100% predictability means 0% innovation*. The blast radius was small enough to absorb experiments nobody could predict.

## Limits: there was no model to copy

Up to here is the story Spotify told about itself. As it spread, the story picked up the name the Spotify model, and countless companies renamed their teams squads and their departments tribes. Yet the first people to doubt the model were the very people who had written the story.

In a 2015 blog post, [No, I didn't invent the Spotify model](https://blog.crisp.se/2015/06/07/henrikkniberg/no-i-didnt-invent-the-spotify-model), Kniberg said he was *just the messenger*. He wrote that although the paper and the videos had become known as the Spotify model, *it wasn't actually intended to be a generic framework or 'model' at all*, only an example of how one company worked. In effect he was underlining the disclaimer on the paper's first page all over again.

More direct criticism came in 2020. Jeremiah Lee, who joined Spotify as a product manager in 2017, wrote in [Spotify's Failed #SquadGoals](https://www.jeremiahlee.com/posts/failed-squad-goals/) that *Spotify doesn't use 'the Spotify model' and neither should you*. Lee wrote that the squad model had always been an aspiration that was never fully put into practice, while leaders were drifting little by little towards more traditional management structures. The post carries weight because it gathers testimony from people who were there. Joakim Sundén, Spotify's agile coach from 2011 to 2017, put it this way: *Even at the time we wrote it, we weren't doing it. It was part ambition, part approximation.*

The problems Lee pointed to are specific. One is the obsession with autonomy. Spotify had no common process for how teams worked together, and since every team worked differently, each collaboration meant figuring out afresh how to work with that particular team. According to Lee, Ivarsson said the series had originally been planned in several parts, but only the part on autonomy ever came out, and the parts on alignment and accountability were never finished.

Sundén's words cut deepest here. *If I were to do one thing differently, I would say we should not be focusing so much on autonomy.* Every new team had to reinvent its way of working from the wheel up, so he thought it would have been better to lay down a minimal common way of working as the default and let teams opt out if they wanted. In terms of the 2×2 from earlier, people remembered only the autonomy in the top right square and mostly forgot the alignment holding it up. And the writing on that alignment never actually got written.

The price of autonomy showed up on the technical side too. The *very little standardization* that the 2014 video noted so calmly became a fairly expensive bill a few years later. A 2020 post on the Spotify Engineering blog, [How We Use Golden Paths to Solve Fragmentation in Our Software Ecosystem](https://engineering.atspotify.com/2020/08/how-we-use-golden-paths-to-solve-fragmentation-in-our-software-ecosystem), admits that autonomous teams had fragmented the ecosystem of development tools, and that the only way to find out how to do something was to ask a colleague. They fondly called this state *Rumour-driven development*. It might work at a startup, but the bigger the company grew, the more it slowed development down. The fix was golden paths. The work of building and deploying a service was laid out as one path the company recommends, with tutorials attached.

The vessel for that path is the internal developer portal [Backstage](https://backstage.io/blog/2020/03/16/announcing-backstage). Onboarding time for new engineers, measured as the time to their tenth pull request, had fallen by 55%.

![Two panels separated by a dotted line. On the left, 'Before, rumour-driven development': teal dashed paths run out of four squad boxes, cross over one another and lead off separately to Tool A, B, C and D, with a note that each squad picked its own tools so the ecosystem fragmented, and that the 2014 video also said there was very little standardization. On the right, 'After (2020), golden path + Backstage': solid orange lines from four squads converge on one point and become a thick arrow, 'One golden path', with a note that the first, for backend, began at hack week. Below that is the company's scale at the time Backstage was released (March 2020): 280+ engineering teams, 2,000+ backend services, 4,000+ data pipelines, and 55% less onboarding time to the tenth PR.](/images/posts/spotify-startup-spirit/rumour-to-golden-path-en.svg)

*How the forks that autonomy created were gathered into one golden path. The numbers come from the 2020 Backstage announcement.*

The list of values changed as well. The Band Manifesto's five values survive in web archive copies from both July 2021 and March 2026, but going to that address today takes you to the [Life at Spotify](https://www.lifeatspotify.com) front page. The front page shows three values under the name The Bassline. One Team starts with the sentence *We believe in the power of alignment*. Human Judgment talks about thoughtful risk-taking and learning from mistakes. I couldn't find an official announcement saying exactly when, or with what explanation, the values changed. All I can confirm is that the old manifesto was still up until the end of March 2026.

![A horizontal timeline with ticks from 2021 to 2026. A solid grey line above the axis runs from 29 July 2021 to 28 March 2026, labelled 'web archive', and its left end connects to a card with the Band Manifesto's five values: Innovative, Sincere, Passionate, Collaborative, Playful. After the solid line ends, a hatched empty stretch runs up to a orange dot at November 2026, labelled 'date of change and announcement not found'. The orange dot connects to a card with The Bassline's three values, One Team, Human Judgment and Make It Happen, marked 'lifeatspotify.com, Nov 2026'. Below the axis, Ek's two layoff memos hang from teal ticks: January 2023, about 6% laid off (too ambitious investing ahead of revenue growth), and December 2023, about 17% laid off (more productive, but less efficient).](/images/posts/spotify-startup-spirit/values-then-now-en.svg)

*From five values to three. The two memos in the middle came in between.*

What stands out to me in this change is how the words moved. The first value of the old manifesto was Innovative, moving fast and taking big risks. The first sentence of the current values is about the power of alignment, and risk-taking now comes with the word thoughtful in front of it. In terms of the 2014 2×2, Spotify never left the top right square, but within it, the company has moved a good way towards alignment.

The Spotify model never existed as written, not even at Spotify. The people who wrote it called it an approximation mixed with ambition, and a snapshot, and the company kept changing its structure, its tools and even its values afterwards. So bringing in the names squad and tribe is a lot like using one photo taken on a day in 2012 as a map. What's more, that photo shows only autonomy, with the alignment that supported it half out of frame. We also have to accept that a good part of the reputation as an innovative startup rested on a story that ran ahead of reality.

### Spotify today

So what does Spotify look like after that photo? Reading its recent engineering blog, the path laid down to make up for the side effects of autonomy is now the foundation for something else.

According to a June 2026 post, [Coding Is No Longer the Constraint](https://engineering.atspotify.com/2026/6/code-with-claude-coding-is-no-longer-the-constraint), more than 99% of engineers use AI coding tools every week, and pull request frequency has gone up 76%. But what the post names as the precondition for this success is consistency, not model performance. Agents work far better when there is plenty of code to refer to and that code is broadly consistent, and in a fragmented codebase their performance drops noticeably, it says. And it points to Backstage as where that consistency began.

Here is my reading. In 2014 Spotify widened autonomy and accepted having little standardization. In 2020 it filled in the rumour-driven development that resulted with a form of alignment, golden paths and Backstage. And in 2026 that same alignment became the ground agents could work on. The square called aligned autonomy is the same as ever, but the centre of gravity inside it has moved over time. It is hard to believe this order was planned from the start. It is closer to the record to say it is what came of retaking the snapshot again and again.

## Take away the disclaimer before the org chart

If I had to boil down why Spotify was called an innovative startup to one thing, I would say it gave its teams a great deal of autonomy while also building the machinery that kept that autonomy from becoming dangerous. The name squad was a label stuck on top of that machinery.

The machinery wasn't perfect either. The writing on alignment never got written, autonomy was paid for in fragmented tools and work around the work, and the company went through two rounds of layoffs and a change of values. But the paper's first page had already warned of this. It said that by the time you read it, things would have changed.

So if there is one thing my team should take from Spotify, I think it is that disclaimer, before any org chart. Treat the way we work now as one snapshot, too, and retake it from time to time to see what has changed. Writing goals down as numbers and checking how far a failure can spread are items to review each time that snapshot is taken. Just as The Playlist pushes back against the previous episode in each new one, Spotify's records rewrote the earlier records every few years. As I see it, that habit of rewriting is the most useful thing Spotify left behind.

## References

**The drama and the book**

- [Wikipedia — The Playlist (TV series)](https://en.wikipedia.org/wiki/The_Playlist_(TV_series))
- [Netflix — The Playlist Launches on Netflix on October 13 (13 Sep 2022)](https://about.netflix.com/en/news/the-playlist-launches-on-netflix-on-october-13-watch-the-teaser-here)
- [MovieZine — Vad är sant i The Playlist? (Swedish, 14 Oct 2022)](https://www.moviezine.se/nyheter/vad-ar-sant-i-the-playlist-gizem-erdogan-ulf-stenberg)
- [Bonnier Rights — Jonas Leijonhufvud, Sven Carlsson, Spotify Untold (original Spotify inifrån, 2019)](https://bonnierrights.se/work/spotify-untold/)

**Technology**

- [Gunnar Kreitz, Fredrik Niemelä — Spotify – Large Scale, Low Latency, P2P Music-on-Demand Streaming (IEEE P2P 2010)](https://people.cs.umass.edu/~phillipa/CSE390/spotify-p2p10.pdf)
- [Fortune — Brainstorm Tech video: Spotify's Daniel Ek on transforming digital music (21 Jul 2011)](https://fortune.com/2011/07/21/brainstorm-tech-video-spotifys-daniel-ek-on-transforming-digital-music)
- [TorrentFreak — Spotify Starts Shutting Down Its Massive P2P Network (16 Apr 2014)](https://torrentfreak.com/spotify-starts-shutting-down-its-massive-p2p-network-140416/)
- [Backstage — Announcing Backstage (16 Mar 2020)](https://backstage.io/blog/2020/03/16/announcing-backstage)
- [Spotify Engineering — How We Use Golden Paths to Solve Fragmentation in Our Software Ecosystem (17 Aug 2020)](https://engineering.atspotify.com/2020/08/how-we-use-golden-paths-to-solve-fragmentation-in-our-software-ecosystem)

**Culture: first-hand sources**

- [Henrik Kniberg, Anders Ivarsson — Scaling Agile @ Spotify with Tribes, Squads, Chapters & Guilds (Oct 2012)](https://blog.crisp.se/wp-content/uploads/2012/11/SpotifyScaling.pdf)
- [Spotify Engineering — Spotify Engineering Culture (Part 1) (27 Mar 2014)](https://engineering.atspotify.com/2014/03/spotify-engineering-culture-part-1), [YouTube](https://www.youtube.com/watch?v=Yvfz4HGtoPc)
- [Spotify Engineering — Spotify Engineering Culture (Part 2) (20 Sep 2014)](https://engineering.atspotify.com/2014/09/spotify-engineering-culture-part-2), [YouTube](https://www.youtube.com/watch?v=vOt4BbWLWQw)
- [Henrik Kniberg — Spotify Rhythm: how we get aligned (8 Jun 2016)](https://blog.crisp.se/2016/06/08/henrikkniberg/spotify-rhythm)
- [Life at Spotify — The Band Manifesto (web archive copy, 28 Mar 2026)](https://web.archive.org/web/20260328031341/https://www.lifeatspotify.com/being-here/the-band-manifesto)
- [Life at Spotify — Our 3 Values: The Bassline](https://www.lifeatspotify.com)

**Criticism and change**

- [Henrik Kniberg — No, I didn't invent the Spotify model (7 Jun 2015)](https://blog.crisp.se/2015/06/07/henrikkniberg/no-i-didnt-invent-the-spotify-model)
- [Jeremiah Lee — Spotify's Failed #SquadGoals (19 Apr 2020)](https://www.jeremiahlee.com/posts/failed-squad-goals/)
- [Spotify Newsroom — An Update on January 2023 Organizational Changes (23 Jan 2023)](https://newsroom.spotify.com/2023-01-23/an-update-on-january-2023-organizational-changes/)
- [Spotify Newsroom — An Update on December 2023 Organizational Changes (4 Dec 2023)](https://newsroom.spotify.com/2023-12-04/an-update-on-december-2023-organizational-changes/)
- [Spotify Engineering — 1,500+ PRs Later: Spotify's Journey with Our Background Coding Agent (Honk, Part 1) (6 Nov 2025)](https://engineering.atspotify.com/2025/11/spotifys-background-coding-agent-part-1)
- [Spotify Engineering — Coding Is No Longer the Constraint: Scaling Developer Experience to Teams and Agents at Spotify (3 Jun 2026)](https://engineering.atspotify.com/2026/6/code-with-claude-coding-is-no-longer-the-constraint)

**GeekNews (Korean)**

- [Spotify's squad team model was a failure (Korean, 1 Jun 2020)](https://news.hada.io/topic?id=2191)
