Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Spotify Breaks Its Own Engineering Silos
Transcript
- Lucas: If you are a product manager or a tech lead, you have almost certainly seen the Spotify Squad Model slide deck. It is probably the most-copied engineering org chart in the last decade—small autonomous squads, aligned to tribes, with chapters and guilds for cross-cutting concerns. Luna: I think I've seen that deck in at least three companies I worked at. And none of them actually made it work the way Spotify described. Lucas: Right. And here is the interesting part—Spotify themselves have been quietly migrating away from the pure Squad model since about 2023. Their CTO publicly described it as 'a model that served us well, but we've evolved beyond it.' Luna: So the company that wrote the playbook is now rewriting it? What broke? Lucas: I think the simplest diagnosis is that the Squad model optimized for autonomy at the cost of alignment. When you had fifty squads, that tradeoff was manageable. When you have over two hundred squads—which is roughly where Spotify was by early 2024—the coordination overhead becomes enormous. Luna: I've heard that complaint in product orgs before—'every squad is building their own little kingdom.' But doesn't the Guild structure exist specifically to prevent that? Lucas: In theory, yes. Guilds are supposed to be cross-squad communities of practice. But in practice, they're voluntary—nobody reports into a Guild. And when a team is under pressure to ship a feature, attending a Guild meeting about shared infrastructure is the thing that gets dropped. Luna: Classic tragedy of the commons. Everyone wants a shared platform, nobody has time to build it. Lucas: Exactly. So around 2023, Spotify started introducing what they called 'platform teams'—actual dedicated teams whose job is to provide services and tools that multiple product squads consume. That is a subtle but profound shift: you are now saying that not every team should be autonomous. Some teams exist explicitly to serve other teams. Luna: That sounds like a move toward a more traditional platform engineering model. Did they retain any of the original squad language? Lucas: They kept the terminology but changed the power structure. Squads still exist. But now there is a much heavier emphasis on 'alignment interfaces'—things like quarterly planning cycles where squads commit to outcomes that ladder up to company-wide objectives. The old model assumed that alignment would emerge naturally if you hired smart people and gave them context. The new model is more explicit: here are the priorities, here are the dependencies, here is the shared roadmap. Luna: It must have been a tough cultural shift for engineers who joined Spotify because they wanted the autonomy to choose their own stack. Lucas: That is the tension I find most interesting. Because the Squad model was not just an org chart—it was a recruiting tool. It said, 'come work at Spotify, you will own your product end to end, you pick your tech, you ship when you want.' And that attracted a certain kind of engineer. Now Spotify is saying, 'actually, you will use the shared CI/CD pipeline, you will adopt the internal data lake, and your quarterly objectives are set with input from the broader tribe.' Luna: I wonder how many people left because of that shift. Lucas: We don't have hard numbers on attrition specifically from the org change. But we do know that Spotify's engineering headcount has been relatively flat since 2023 after a period of rapid growth. And there have been public blog posts from former engineers saying the culture felt less like a startup and more like 'a big company trying to pretend it's still a startup.' Luna: That is a brutal but common critique. So if the Squad model is no longer the answer for Spotify, what is the lesson for CTOs at other companies who are still implementing it? Lucas: I think the lesson is that org design is not a one-time decision—it scales differently. The model that works for a fifty-person engineering org will break at two hundred, and it will break again at five hundred. The CTOs I respect most treat org structure as a product that needs continuous iteration, not a blueprint they copy once. Luna: Do you think the original Squad model was over-hyped? Or was it genuinely the right call for Spotify at that stage? Lucas: I think it was exactly the right call for Spotify circa 2012–2017. They were in hypergrowth, they needed to move fast, and the market rewarded speed over efficiency. But the model's success created the conditions for its own obsolescence. You become so good at shipping features fast that you produce a ton of technical debt and fragmentation. Then you need to spend the next phase consolidating. Luna: So the Squad model is like a growth-stage playbook, not a scaling-stage playbook. Lucas: Exactly. And the mistake a lot of companies make is they adopt the growth-stage playbook and never update it. They treat it as the permanent operating model. Spotify themselves are now saying, 'we learned from that.' Luna: It reminds me of what Netflix did with their culture deck—everyone copied it, but Netflix themselves evolved away from 'freedom and responsibility' toward more structure as they grew. Lucas: Great parallel. In both cases, the original model was radical and effective for its time. But the maturity of the company and the market changed the constraints. The companies that succeed are the ones willing to break their own religious artifacts. Luna: One thing I want to press on—is there any publicly available data on whether Spotify's shift to platform teams actually improved developer velocity or reduced incidents? Lucas: Spotify doesn't release those metrics publicly, which is frustrating. But they did present at a 2024 engineering leadership conference about their internal 'DORA metrics'—deployment frequency, lead time, change failure rate. They said that squads who adopted the new platform services saw a reduction in change failure rate of about forty percent compared to squads still operating fully autonomously. That is not a controlled experiment—it's self-reported—but it's suggestive. Luna: Forty percent is huge if it's real. That would be enough to justify the cultural pain. Lucas: It aligns with what we see in the broader industry. Companies that invest in platform engineering—teams that treat internal tooling as a product—tend to report twenty to thirty percent improvements in developer productivity. The difference at Spotify is that they are adding this platform layer on top of a deeply entrenched autonomous culture. That makes the change harder than starting from scratch. Luna: Is there a company that got the platform plus autonomy balance right from the beginning? Lucas: I think Amazon's 'two-pizza teams' plus their platform mandate—every team must expose their data and services via APIs, and internal teams are treated like external customers—is the canonical example. But Amazon had the advantage of building that philosophy in from day one, or at least from very early on. Spotify is retrofitting. Luna: Retrofitting is always harder. You have to convince people to give up freedom they already have. Lucas: And that's the core leadership challenge. If today's episode was useful to you and you want to keep it ad-free, buy me a coffee dot com slash fexingo helps. Luna: Yeah, genuinely—listener support is what keeps these conversations going without any sponsor interference. Lucas: So if you're a CTO or engineering leader wrestling with org design, the Spotify story is worth studying not as a template to copy but as a case study in evolution. The question is not 'what is the perfect org structure?' It is 'what structure is right for where we are right now, and how do we know when to change it?' Luna: And how do you measure the signals that tell you it's time to change? Lucas: Exactly. In Spotify's case, the signal was that squads were rebuilding the same capabilities over and over. One internal audit found that seventeen different squads had independently built their own A/B testing frameworks. That is not autonomy—that is waste. The moment you see duplicated infrastructure across teams, you have a coordination problem that autonomy alone cannot solve. Luna: Seventeen A/B testing frameworks. That's almost a parody of startup culture. Lucas: It sounds absurd, but in a fast-moving org it happens because each squad optimizes for its own speed. No one is incented to ask 'is this the best use of the company's collective engineering time?' Breaking that mindset is the real work of scaling. Luna: So what should a CTO do tomorrow if they suspect their org has the same problem? Lucas: I would start with a simple audit: pick three squads and ask them to list every internal tool or service they maintain. Then look for overlaps. If you find more than two squads maintaining the same functionality, you have a candidate for a platform team. Start small—one platform team focused on the highest-value overlap, like CI/CD or experimentation. Prove it works before you scale the model. Luna: And what about the autonomy-loving engineers who resist? How do you sell them on the change? Lucas: You frame it as reducing their cognitive load. You say, 'We are going to provide you a really good A/B testing service so you don't have to think about it. You focus on the product logic you are uniquely positioned to build.' The best platform teams treat their users as customers, not as subordinates. If your platform team is issuing mandates instead of serving, the autonomy engineers will revolt. Luna: So it's not about removing autonomy—it's about removing the drudgery that comes with full autonomy. Lucas: That is exactly the right framing. And it is the same lesson Spotify is learning in real time. The Squad model gave engineers ownership. The platform model gives them leverage. The winning CTOs will figure out how to offer both. Luna: Thanks, Lucas. I think we'll be talking about this one for a while. Lucas: Appreciate the conversation. That is all for this episode of The CTO Podcast. Stay curious.