We all know the Spotify Model. But what does it actually look like in practice? In this episode of Ship Happens, Pia Nilsson from Spotify breaks down how autonomy, accountability, and standardization work together to power engineering at scale. From squads and synchronization to platform tooling and AI adoption, this conversation explores how one of the world’s most talked-about engineering cultures continues to evolve.
The Spotify Model has become one of the most referenced—and misunderstood—frameworks in modern software development.
In this episode of Ship Happens, we sit down with Pia Nilsson of Spotify to explore what the model really means in practice. Beyond the buzzwords, Pia explains how Spotify balances creative autonomy with clear standards, cross-team synchronization, and operational accountability.
The conversation dives into Spotify’s platform engineering evolution, including how internal tooling like Backstage helps reduce toil and improve developer effectiveness. Pia also shares how standardization, context-sharing, and intentional time for innovation—like hack weeks—have shaped engineering productivity across the organization.
We discuss:
Whether you're leading engineering teams, building internal platforms, or scaling organizational structures, this episode offers a grounded look at how squads, standards, and scale coexist in a real-world, high-performing tech organization.
Hit play to go inside Spotify’s engineering culture.
Pia Nilsson is a leader at Spotify focused on engineering culture, platform evolution, and developer effectiveness. She has played a key role in shaping how Spotify balances autonomous squads with shared standards, enabling teams to innovate quickly while maintaining alignment across a global organization.
(00:00) The hidden cost of “special stack” productivity — and why agents need order
(00:39) Welcome and meet Pia Nilsson (Spotify Platform DevX & Backstage)
(01:30) The real Spotify Model: autonomy for innovation
(03:48) Where the model breaks at scale
(04:53) Alignment at scale: OKRs, company bets, and synchronization
(08:12) Translating DevX into leadership KPIs
(10:20) The 2017 wake-up call: 60+ day onboarding
(11:19) Backstage and golden paths: reducing fragmentation
(13:48) Removing toil: fleet management and automation
(15:36) Creating space for innovation and hack weeks
(18:51) AI at Spotify: trials, metrics, and real productivity gains
(20:52) Can you copy Spotify’s culture? Core principles explained
(24:20) Backstage deep dive: onboarding and portal UX
(25:37) Backstage as a single pane of glass
(26:46) Beyond the web UI: IDE, CLI, Slack, and shared context
(29:00) Will AI replace rich UIs? Trust and visibility challenges
(32:26) Why Spotify open-sourced Backstage
(34:51) From open source to commercial product
(36:01) Product prioritization: internal vs. external customers
(39:29) AI-ready engineering: standards, skills, and orchestration
(43:28) New bottlenecks: PR review, security, and testing
(47:53) What’s next: AI-powered fleet management and wrap-up
Squads, Standards, and Scale: Pia Nilsson on Engineering at Spotify
[00:00:00] Pia Nilsson: You feel really productive having your special stack that you love. Nobody else understands what you're doing. And, uh, you can't move teams now because team can't work without you. So what we have been discovering is that the agent actually needs. What the humans need. Order coding agents are great and everyone loves them, but it's, it's just pushing the bottleneck further down.
And the whole industry knows and sees this. I think we certainly feel it as well. And right now the bottleneck is of course the PR reviews and then you can, of course leverage agents to reduce to as well, so. It all sort of, uh, builds on each other.
[00:00:39] Per Krogslund: Welcome to Shape Happens, a podcast where I sit down with some of the smartest people in the industry to talk about the skills, techniques, and mindset that helps you shape amazing software.
So today I'm talking to Pia Nelson, senior Director of engineering at Spotify, working on the platform developer experience and as GM of. Backstage. I used to work with Pierre at Spotify and she's one of the top of my guest [00:01:00] lists. Um, as we started planning this session, Pierre has years of experience and develop of experience, strong opinions about how you scale engineering and how you condense years of learning into pro a product like backstage.
We have plenty of things to talk about, so I think we should just get started and it's a big pleasure to welcome you to the show, Pierre.
[00:01:28] Pia Nilsson: Thanks for having me. I'm super excited.
[00:01:30] Per Krogslund: Alright, so, um, I really wanna start with just setting the stage and just talking about one of the most misunderstood ideas in tech over the last decade, which is the Spotify model.
I think. Uh, it was touted as like revolutionary new way of working that you went, invented at Spotify and used yourself. And I think everyone from McKinsey or every consultant out there has kind of like said this is the way to work. And, um, I've also worked in, in a lot of different companies where like pieces of this model [00:02:00] has been implemented.
I would say poly or maybe just like partially. And it usually started by just renaming teams into squads. Uh, so, um. So j but just like high level, what is the philosophy behind the Spotify model?
[00:02:16] Pia Nilsson: Yeah, it's, uh, certainly been misunderstood and misinterpreted a lot, uh, over the years. So the, the core idea is, uh, to empower people.
[00:02:28] Per Krogslund: Yeah.
[00:02:28] Pia Nilsson: Because the, we really believe in the power of creative people. And people who feel ownership, uh, are gonna be more creative. So there's a lot of autonomy built into the model. Hmm. And that is why it's not autonomy for its sake or for the sake of autonomy. It's for the sake of innovation.
[00:02:49] Per Krogslund: Yeah.
[00:02:49] Pia Nilsson: And also for sort of believing in people.
So there's a real sort of, um, message of love, I would say, to human beings, uh, uh, built [00:03:00] into the model that's the core of it.
[00:03:01] Per Krogslund: Okay, but it's not all about autonomy, right? It's not like people can't just do whatever they want. Like it's not like, yeah, we are gonna just sit back or we are just gonna do this thing or whatever.
Like, there's also accountability, responsibility, alignment, and so on.
[00:03:17] Pia Nilsson: Yes. So, so true. And, uh, it's not that we, uh, didn't mess it up as well. It started out I think, really, really well. Uh, uh, in the first years, um, before my time at Spotify. But when I joined we were like, uh, around 1000 to 2000 engineers or something a long time ago.
Now we're, uh, a bit larger than that. Uh, we had started to run into a bunch of challenges that I think many others do as well. So. It, it has to do with all of this sort of, what, what is the impact of what we're building. So the, the model really tried to, uh, uh, help people to get [00:04:00] to the right impact through enabling them with.
Ownership of a domain.
[00:04:05] Per Krogslund: Hmm.
[00:04:05] Pia Nilsson: However, when as you grow, then uh, the domains become so large, you're gonna have to have several teams owning that huge domain.
[00:04:14] Per Krogslund: Yeah.
[00:04:15] Pia Nilsson: And then, uh, synchronization wasn't built into the model.
[00:04:19] Per Krogslund: Okay. So
[00:04:20] Pia Nilsson: that's sort of when it started to break down for Spotify, in my view, where,
[00:04:24] Per Krogslund: okay,
[00:04:24] Pia Nilsson: the domains were so big, the company's, uh, needs were so huge.
Uh, we were 50 to 100 people working on a, a huge problem of sort of the backend engineering, for example, improving, improving the backend engineering, uh, for all of the thousands of developers. That's not the one team problem any longer. So then the synchronization is massively needed and we sort of needed to add onto the model in order to.
To enable people to also collaborate a whole lot more.
[00:04:53] Per Krogslund: Yeah. So when you say synchronization, what does that like specifically mean? In like the case of Spotify,
[00:04:58] Pia Nilsson: my, uh, [00:05:00] manager, Tyson Singer, he, uh, he calls it uh, uh, synchronized swimming because he wants to stir up some. Annoy some folks. Yeah. Because of course, that's super silly, but it, and it's sort of, it's very counter to, uh, US engineers who thinks we are, we are so incredibly smart.
Uh, but we also need to sort of just do the, do the job of synchronizing teams across teams. Uh, teams of teams.
[00:05:26] Per Krogslund: Yeah.
[00:05:27] Pia Nilsson: So, and. What we, what we really believe in there is sort of, uh, we have used different KPI models. We use o OKRs. Right now we're using, uh, something called the company pets. And in order to make this whole thing work as the whole r and d to actually, uh, work effectively synchronized.
Uh, as much as possible when you're several thousands of people, uh, it has to start at the top. So what the leadership at Spotify did, um, I would say [00:06:00] three years ago now, uh, they collapsed all of the leadership teams at the highest level into one team of 18 people-ish, which is pretty cool. Yeah. So, and they meet three hours a week.
So you can imagine that that kind of time for that level of leader is pretty massive.
[00:06:19] Per Krogslund: Mm-hmm.
[00:06:19] Pia Nilsson: What happens? Since then, it's sort of the clarity cascades down and it so much work is, is, uh, just clarified, maybe never done.
[00:06:30] Per Krogslund: Yeah.
[00:06:31] Pia Nilsson: Because it's not prioritized. And now we know because those two leads have, have, they have the same priority list.
There's just one and it's called the company bad list.
[00:06:41] Per Krogslund: Yeah.
[00:06:42] Pia Nilsson: So there was a lot of work going into that. A great planning team, of course. Really, it starts at the top. It's really quite simple, you know, but, uh, to, to pull it off with, uh, uh, several thousands of people, then that's a bit of a challenge, of course.
But that's how it starts. And then it trickles down. [00:07:00] And then you can really have, uh, teams that own something particular that is actually prioritized and staffed, and they can then feel ownership.
[00:07:10] Per Krogslund: Mm-hmm.
[00:07:11] Pia Nilsson: And, uh, the innovation can start. So. In my view, uh, sort of top down in this a little bit, actually created far more empowerment.
[00:07:22] Per Krogslund: Yeah.
[00:07:22] Pia Nilsson: In, in where, where I sit, where the teams that I, I can see they, they know, they work on something really important. They know they're gonna be prioritized and ship at a certain date, so then the creativity really is triggered.
[00:07:38] Per Krogslund: Yeah, and I guess that's, that's, it's funny how you describe it here.
Actually, taking a top down approach really goes against everything. Like a lot of people have publicly said about the, the, the Spotify model. It is actually a lot of like top down control that sets the stage for how do you focus and you know, you're focusing on the right thing and then, then you're free to make the right decisions within that, that sphere.
You, you kind of own right.
[00:07:58] Pia Nilsson: Hmm. [00:08:00] Yeah. And then one, one question I thi I think that comes is sort of, well, will it only be prioritized by the business needs? What about the platform needs? But I think, yeah, we're gonna get in, get ourselves into that soon.
[00:08:11] Per Krogslund: Yeah. Yeah. I was about to ask, like you are, you're, you're sitting in the platform organization, right?
There's a, and I know for my time as perfect, it goes a lot of, a lot of work into a platform, uh, to enable this type of work. So, um, chip, let's, let's dive into that really. Like what are your key levels levels as like a platform organization to really enable how, like people at Spotify work?
[00:08:33] Pia Nilsson: Oh, we are, um, we have a lot of sort of, um, we have.
A lot of help from, uh, leadership being very technical at Spotify. Mm-hmm.
[00:08:42] Per Krogslund: Yeah.
[00:08:42] Pia Nilsson: So, uh, the, the key thing for us is to enable to translate what we're doing on the platform level into, uh, understandable KPIs at the top level.
[00:08:55] Per Krogslund: Yeah.
[00:08:55] Pia Nilsson: So we are talking about sort of developer effectiveness. Of course in, [00:09:00] in with the Dora metrics, but also some other metrics that we are, we are leaning on quite a bit.
Um, meantime to change, for example. That is one of our favorite metrics. Uh, but because we wanna sort of, that speaks clearly to leadership, uh, something they really care about. And then we can also show did this initiative that we ran. Actually improve it.
[00:09:23] Per Krogslund: Yeah.
[00:09:23] Pia Nilsson: We can also hold ourselves accountable, which is incredibly powerful.
Of course, building a lot of trust. So I think it's, it's a lot about sort of translating our work into understandable KPIs, uh, that are, um, that aren't at the aggregate level, that you can't see any difference regardless of what you're doing. Almost. They need to be, uh, at a low level, low enough level so that you can actually see the difference for a platform initiative.
[00:09:57] Per Krogslund: Uh, and, and so [00:10:00] I think how, how has, like what is the most important things you've done to kind of see that? What's, what's been the most effectful things you've done as a platform over the years? Like, what is you, you probably tried a bunch of stuff that, me, kind of me, but like what's your, what's your like greatest hits of like, platform engineering that really made a difference?
[00:10:17] Pia Nilsson: Oh, wow. So I think, I think, um. We ran into this massive problem back in 2017 where, where we were still hiring, uh, very rapidly. And we were slowing down, uh, we could see that in our metrics that we were actually slowing down. Uh, the onboarding time for, for, for a Spotify engineer, uh, was over 60 days.
[00:10:43] Per Krogslund: Oh wow.
Yeah.
[00:10:43] Pia Nilsson: Uh, and we had the target of being 20 days, so we were no way near and we were going the wrong day way as well. Um, so we were like super worried, uh, where that the more people we hire, we were hiring actually the slower we become, which is [00:11:00] like terrifying Of course.
[00:11:01] Per Krogslund: Yeah.
[00:11:01] Pia Nilsson: Because we really needed to hire those people.
Uh, so that was a clear and, and, uh, exact enough metric to, to reason around for leadership, uh, as well as for us. Uh, so we could really try stuff out to, uh, improve it. So,
[00:11:19] Per Krogslund: Hmm.
[00:11:19] Pia Nilsson: Sort of outta that came a whole bunch of, of things that I'm, uh, super proud of being. Part of creating. One is, of course unifying all of the tools and stuff, uh, behind our development portal backstage.
Uh, that really, uh, was speeding. Speeding folks up. Uh, yeah, hugely. The, the other thing is we had, uh, we, we finally took the step in our very autonomous engineering culture to set our foot down and like, we're gonna have standards. And you per, you were, you were there when we created all these golden tracks.
[00:11:53] Per Krogslund: Yeah.
[00:11:54] Pia Nilsson: So there was a lot of conversations with the engineering, uh, [00:12:00] population of course, because we really care about what they think should be golden. Yeah,
[00:12:04] Per Krogslund: yeah.
[00:12:04] Pia Nilsson: As a golden standard for backend. What should a backend service look like? Mm-hmm.
[00:12:08] Per Krogslund: Did you meet pushback on this like, standard thing of developers feeling they couldn't be creative anymore or like
[00:12:14] Pia Nilsson: Yeah, and there were, there was that, which was reported and there were also the tabs versus spaces kind of conversations.
Oh. You know, so there were so many, like a range of, uh, uh, especially Spotify engineers at least mm-hmm. Did not want to have any kind of autonomy, uh, uh, pulled out of their, like under, from, under, under them. So they really, uh, fought this for some time, but then we could really share metrics and people really relate to no, everyone wants to do good.
Yeah. They were all good citizens. Yeah. And when they could see we're slowing down folks, this isn't working. Yes. You feel really productive. Mm-hmm. Having your special stack that you love. Nobody else understands what you're doing. [00:13:00] And, uh, you can't move teams now because. Team can't work without you.
[00:13:05] Per Krogslund: Yeah.
[00:13:05] Pia Nilsson: So it, it's sort of, we had so many examples of, uh, this is not effective people. So we kind of over time won them over all of them and sort of now we have, and you know this, so well pair we like over the years, we are now at the 75 ish Golden State hit rate that we're striving for all the time. So standardization, I think is the, is the second or backstage as in unifying the whole ecosystem.
And uh, then standardization. Is through Golden States we call them. And Golden tracks. And Golden Paths and everything. Golden.
[00:13:42] Per Krogslund: Yeah.
[00:13:43] Pia Nilsson: But that's, I think is a huge impact that the platform had for Spotify as well.
[00:13:48] Per Krogslund: Yeah.
[00:13:48] Pia Nilsson: And then I think removing toil is also very important. Not just, you know, guard railing with standards, but we, you need to.
Remove that toil, uh, as [00:14:00] so many of, uh, people are saying in the industry right now. Uh, and it's the, the fragmentation of all of the solutions, uh, makes it pretty difficult to remove all that, all of that toil. Yeah. It also makes it pretty difficult for agents to reason without hallucinating. If you have so many different versions of a solution, what, what is the agent gonna do?
So what we have been discovering is that the agent actually needs what the humans need order.
[00:14:30] Per Krogslund: Yeah.
[00:14:30] Pia Nilsson: In order to go fast and not be incorrect. So standards really help humans and agents. And then reducing the toil is possible in automated fashion if you have the standards. So it's really building on each other.
And then you can, of course, leverage agents to reduce toil as well. So, yeah. It all sort of, uh, builds on each other. But we built this thing, uh, called fleet management where we really, [00:15:00] uh, removed bunch of toil for so today, which is like bot, bot generated code. So all of our security updates and all of our version upgrades and all of, all of the lower level, uh, tasks that engineers hate doing because they're pushed onto them basically.
[00:15:16] Per Krogslund: Yeah.
[00:15:17] Pia Nilsson: That is being done by, uh, our bots since, I don't know, two and a half years back. Yeah. So 50% of all written code at Spotify is already bot generated, which is like, I think that is, that should be up there on the top three.
[00:15:33] Per Krogslund: That is,
[00:15:33] Pia Nilsson: that
[00:15:33] Per Krogslund: is like Go, go for it. Go for it.
[00:15:35] Pia Nilsson: Sorry. Uh, I just wanted to really talk about the human aspect because it's, it's such a big important, uh.
Thing to really strive for, um, which is innovation.
[00:15:51] Per Krogslund: Mm.
[00:15:51] Pia Nilsson: Everyone needs innovation. How do we help people to innovate, also to innovate about the right [00:16:00] things?
[00:16:00] Per Krogslund: Yeah.
[00:16:01] Pia Nilsson: So it has to do with transparency, culture that I was talking about before. That's key because you need the people to innovate about the right problems.
[00:16:09] Per Krogslund: Yeah.
[00:16:11] Pia Nilsson: And then you need the standards so that the innovation not just, uh, creates more fragmentation. And then you need to remove that toil so that people actually have some time. But that's not enough. People might not, people might just be good citizens and work hard and never have time to really innovate because there's so much work to do all the time.
Yeah. Everyone knows that. So one has to really help them put the time on the cal calendar to innovate, which is of course what hack weeks and hack days are for. So we have the yearly hack week, which is like wholly at Spotify. We are, we are ordered to hack all of us, including myself and my vp. They're all hacking one week.
It's incredible [00:17:00] actually. Uh, and, but that's just one once a year. And then we have, yeah. Uh, quarterly hack weeks for all of the sort of 40 people ish orgs. And those are, of course, I think the most valuable things, uh, that I see that actually translate immediately to fantastic product features.
[00:17:18] Per Krogslund: Mm-hmm.
[00:17:18] Pia Nilsson: Or coming from those, yeah, from those hack weeks because then, you know, it's, it's, um, yeah.
It's, you're gonna work on something that is on your mind and the next hack week is just, you know, a few months out. So it's gonna be something really fresh and something really relevant, and then you're gonna work on it. Uh, you're gonna be, be able to put that inside of the product you are working on anyway.
So it's, it's really, uh, a ous spiral I think.
[00:17:49] Per Krogslund: I think, I think I also remember from the time, I think one of my favorite features in, in the Player came from a hack week, I think it was what's called day List now, which is like these, like autogenerated ones. And I, I, I think I use it [00:18:00] every day just like putting on a day list that just matches whatever I'm used to listening to.
Um, so I, it is, so I think it just, it is a lot of pieces here. I'm just trying to summarize here. You, you, you have, you have standards and golden paths in place that really. Takes the the thinking about like what is the fundamental infrastructure that people use for the different things, and you remove a lot of the toil by using fleet management.
And then you really create space for innovation because yeah, you don't have to think about these things. That's really a, a commodity and you also dedicate a lot of time to just like do other things than in your normal day in day, day-to-day job. And I think also, like one thing that you haven't really talked about here is like Spotify is, is working at tremendous scale.
So I think just the, the economics of scale that you have tens and thousands of services that just. Gets updated every day and just gets their security patches and no one really needs to do anything. It is already, like, that's a tremendous amount of like developer hours. Um, and I guess it also sets you up for, for success in like adopting AI agents at a bigger scale than a lot of other companies who don't have all these things already.[00:19:00]
Like you have the fundamentals in place to really do this at a faster rate than a lot of other companies.
[00:19:06] Pia Nilsson: Yeah, that's
[00:19:07] Per Krogslund: true. Yeah. Yeah.
[00:19:09] Pia Nilsson: Did you want me to chime on that?
[00:19:11] Per Krogslund: Please do, please do.
[00:19:13] Pia Nilsson: I think it's very true, and I think the, the research mentality really helps us. It comes from the autonomous culture and the passion for the domain.
[00:19:22] Per Krogslund: Yeah.
[00:19:22] Pia Nilsson: And the, so the research mentality is really helping us a lot as well in this AI era, uh, sort of assessing the, for example, the AI IDs. Yeah, which one should we pick? Should we pick them all for which use case? So we run sort of trials and we have our developer experience and developer effectiveness metrics.
Uh, so we actually see what mattered for our productivity.
[00:19:50] Per Krogslund: Okay.
[00:19:50] Pia Nilsson: What helped? Really helped, and that's just so exciting to learn for all of us. I think so, I, we, we constantly sort of. [00:20:00] Keep learning about, well, how should we develop code now? Yeah, it's gonna change every month, at least, uh, in, in this, these times.
Certainly.
[00:20:12] Per Krogslund: Yeah. So you actually been trying out different AI agents with different teams and just seeing the impact. So you actually have numbers, you have receipts on actually what works.
[00:20:21] Pia Nilsson: Yes, we do. And that, I just took the example of, um, IDs, but we of course have done the same for, um, coding agents and for agent builders and for a few other things as well.
[00:20:35] Per Krogslund: Yeah. So again, another advantage of just doing all this work that kind of like in this age seems kind of boring, where people are just like shoveling AI into an organization and hoping for the best you actually know what actually works.
[00:20:48] Pia Nilsson: Yes.
[00:20:48] Per Krogslund: So I think we do.
[00:20:49] Pia Nilsson: It's, it's super exciting. Yeah.
[00:20:50] Per Krogslund: Yeah. Amazing.
So I think just, uh, uh, one last question on like your culture, uh, which I want to, I want to start talking a little bit about backstage and the open sourcing out, [00:21:00] but I think just like, um. When you think about this culture and all the things you just mentioned, is this, is it really possible to adopt the user Spotify model or the culture without, you know, the platform practices and all that hard work?
You're talking about enabling this, is it even possible to do
[00:21:18] Pia Nilsson: well? I really think so.
[00:21:19] Per Krogslund: Yeah.
[00:21:20] Pia Nilsson: I really do. Um, because it's not about having, it wa it wasn't like we were so perfect.
[00:21:26] Per Krogslund: No.
[00:21:27] Pia Nilsson: Uh, so we, we were a complete mess. Uh, uh, love lovingly. I'm saying this fun ca creative chaos, uh, back in 2017. And, uh, so adopting the Spotify engineering approach and to some extent maybe culture is simply following a few principles, and I think many, most people can do that if they want to.
So those principles are. Making the right thing the easy thing. [00:22:00]
[00:22:00] Per Krogslund: Yeah.
[00:22:01] Pia Nilsson: So that is sort of talking to the standardization. So we can't just go ahead and top down, say you're gonna, you're gonna now build backend services like, like this.
[00:22:10] Per Krogslund: Hmm.
[00:22:10] Pia Nilsson: You gotta make it attractive. Uh, and also so that the, at Spotify, it was really healthy for us that the engineering org.
It wasn't the captive audience, if you see what I'm saying. Yeah. We couldn't force our solutions on them. We had to make them better so that they were the, the ones people preferred. Maybe not everyone can get there, but it, it was very helpful as a sort of a principle. And then, then another principle I wanted to, uh, I already touched on sort of is taking the pain we call it.
[00:22:45] Per Krogslund: Hmm. So
[00:22:46] Pia Nilsson: then that is reducing the toil. So how we can do, companies can do that in very many ways, you know?
[00:22:52] Per Krogslund: Yeah.
[00:22:53] Pia Nilsson: Uh, so, uh, and then, and then finally you can of course make it to actually [00:23:00] automate a bunch of changes. Uh, but there are many, many steps before that that that is gonna be super helpful for everyone.
Yeah. I think, and I think the, the last principle. And also builds on this, and it's like, move up the stack.
[00:23:16] Per Krogslund: Okay.
[00:23:17] Pia Nilsson: You gotta constantly try to move, move your teams up by keeping them excited about the future. Focusing forward is moving up the stack. It will move them up the stack.
[00:23:29] Per Krogslund: Yeah,
[00:23:31] Pia Nilsson: I think that is, uh, very inspiring, uh, and engaging for teams.
Uh, so one has to figure out a way to, uh, reward. That behavior.
[00:23:43] Per Krogslund: Yeah. And are you, you're thinking about this specifically in, in AI adoption or is more general, just like you as a platform team is really one of your core jobs, is to really move people up the stack and and keep them interested in new technology so they didn't just get stuck in their ways or
[00:23:58] Pia Nilsson: I, I'm, yeah, exactly.
I, I [00:24:00] think it was very relevant before ai, but now it is sort of a must
[00:24:05] Per Krogslund: yeah.
[00:24:06] Pia Nilsson: For people to welcome the changes coming our way. So that we can figure out, well, how do we help our companies now? And that is exciting to me.
[00:24:16] Per Krogslund: Yeah, that's exciting. Alright. So I think with that, um, let's talk about backstage. Um, again, from my experience at, at Spotify, like the eng, the internal engineering platform you had at the time, which I think is mostly what people can get from, from backstage now is like, there was, I'd never seen anything like it, it was just like, this is like an amazing experience to even like come in as a new.
Personally in the organization, you just, you just very quickly knew where everything was. And, and I think your your point about the, the onboarding time going drastically down, um, I see that like, that's, that's a very big part of it. So I think maybe let's, let's talk a little bit about Backstage. What does it do?
Like it's an internal. Portal that really shows engineers, where's all the documentation? [00:25:00] Here's all the services, here's the data. And a bunch of other, you, you have a bunch of internal tools as well, like for experimentation and your, your AI agents and well, a bunch of other things. Maybe we can dive a little bit into that as well, like, um, but I think if you've from, from like as the owner of Backstage, you would say, like what is, like, do you talk about meantime to, to, to change?
Um. And is that your North Star? What you wanna do with, with Backstage?
[00:25:29] Pia Nilsson: Great question. I think just placing backstage in the, um, DevX ecosystem
[00:25:37] Per Krogslund: mm-hmm.
[00:25:37] Pia Nilsson: It has, it sort of solves the problem of synchronized swimming
[00:25:41] Per Krogslund: Yeah.
[00:25:41] Pia Nilsson: In so many different levels. So you gotta transparently share what are we doing, what are we.
As a company.
[00:25:49] Per Krogslund: Yeah.
[00:25:49] Pia Nilsson: That's where you, you need something like backstage. It's super powerful. Um, then there is of course a lot of actions that you can be [00:26:00] tying into all of, all of the plugins as we call them, uh, inside of Backstage. Yeah. And then, then all of those actions will of course be, um. The right actions, the, the approved ones from the platform organizations, which is again, super powerful because otherwise, without this single pane of glass, you would have 20 different solutions or actions, uh, to solve the same problem.
Uh, I think, uh, it's just impossible to synchronize if it's not vis visually visible someplace Now. This is very important for sort of the synchronized swimming, and I started saying this because there is this bigger picture as well that I wanted to share with you. At Spotify, we don't think spot uh, engineers and dev teams only should be in backstage.
Backstage is one developer surface, which is super important for the synchronous swimming, uh, use case. [00:27:00] But in order, like what should a great developer experience look like? Well, actually we believe there are several different developer surfaces that the backstage ecosystem needs to power. And with what?
With context?
[00:27:14] Per Krogslund: Yeah.
[00:27:14] Pia Nilsson: Because at the heart of it, backstage is context. Right. It's documentation. It's the catalog metadata. It's what you call what? Yeah, which is incredibly important. Is this data set critical, business critical, or is it an experimentation or ex experimental? Those are the things that are business critical to know and to have someone accountable for, and all of that is stored.
That is sort of the gold. In backstage, which is of course the AI fancy word context. So Backstage heart needs to power up these other developer services, which is of course the ai IDE or whatever IDE one perverse needs to have an MCP out of the [00:28:00] backstage. The backstage MCP powering it so that when, when you vibe code your next backend service, it's according to your company's standard.
That's all possible. Of course, if you have that, um, using that MCP Yeah, that, that's at least how we believe we one should go. So that's one, one, uh, developer surface. Another developer surface is the CLI. So I think it's gonna come more and more over the years as we're seeing some AI vendors are super powerful, like Claude, uh, people really love working through ACL I.
So this is a very, it has always been a very important interface for developers. And it will just, um, become even more important probably as sort of prompt based engineering, uh, is, is, uh, in the future for us here. So that's in our other, and then, then there are a few more, like Slack is very important for us, for example, and then backstage as a fourth developer service.
[00:28:59] Per Krogslund: Yeah. [00:29:00] So, um, so I think, yeah, with, with, uh, with agents and people getting used to chat and so on, and the CLI, again, like we, it's kind of funny, we are go moving back to this like kind of retro, we are moving a little bit away from the rich interfaces and like into. Back into the terminal window. And so do you feel that this is where like you can say developer experience might be more of in the future, whereas before we really needed like a rich interface to explain how is this service performing?
Are we doing the right things and so on. But just now you have kind of AI as a between layer that can really like get that context and explain it to you or use it with something else. So is, is developer experience moving more into that direction again instead of having these rich interfaces?
[00:29:44] Pia Nilsson: I mean, to some extent I think it is.
[00:29:46] Per Krogslund: Yeah. Uh,
[00:29:47] Pia Nilsson: we are measuring this on a week by week basis to learn what actually is effective for us.
[00:29:53] Per Krogslund: Yeah.
[00:29:54] Pia Nilsson: So I think the true answer here is to be recognizing we are on the learning journey here [00:30:00] for, hmm. Maybe forever.
[00:30:02] Per Krogslund: Yeah.
[00:30:02] Pia Nilsson: So, uh, what, what we know from now by now is that we actually have a, there's a very big use case still for the, the backstage at the web ui, but as I was saying, the heart of it is the knowledge.
[00:30:18] Per Krogslund: Mm-hmm.
[00:30:18] Pia Nilsson: And the decisions. Uh, for the company decisions on, on on our components. Yeah, those are incredibly important and that that's what really, really matters. You know, regardless of where AI takes us, uh, then the surfaces, they are gonna change over time. Right now for us, we are do, we do see a lot of, uh.
Uh, people going to backstage, still the UI and, and, uh, we, we don't see any kind of drop in that. Hmm. So we, we lack the data point that you're referring to, but we are also thinking it, it will have an impact soon.
[00:30:56] Per Krogslund: Yeah.
[00:30:57] Pia Nilsson: What we are seeing is people really need, um. [00:31:00] Not sort of the hallucinated graphs, maybe from some coding agent, which might be right or might not be right if they wanna see the AI costs.
For example. One of the things that people are looking at pretty much in the industry right now, they probably want to go into the cost insights plugin. Where the team has consolidated the data. Mm-hmm. Uh, so that they know I'm prepare, I'm comparing apples to apples here.
[00:31:25] Per Krogslund: Yeah.
[00:31:26] Pia Nilsson: Uh, however you can, you can of course ask through your MCP, uh, uh, backstage, MCP, you can ask that plugin for that same data.
Well, what will it do to it in, in, in between there, sort of how can you trust it? Yeah. So that's, uh, I think there's a lot. We are, for example, leveraging the coding agents best of breed. Uh, we're using all of them, uh, to compare them to each other.
[00:31:53] Per Krogslund: Yeah.
[00:31:54] Pia Nilsson: Uh, and we are finding a lot of success in that. So there are, there are gonna be different use cases where we [00:32:00] leverage the LLMs more and more and soon we're gonna figure out like, how do we leverage them for the UI as well.
[00:32:07] Per Krogslund: Yeah.
[00:32:08] Pia Nilsson: Yeah. We'll have to stay tuned and learn.
[00:32:11] Per Krogslund: Yeah. But right now it is. So, again, just to re reiterate your, reiterate, your point here is like the, the key feature here is really the data and the context and that that allows us to actually build new experiences. That's actually. Stable and actually works and actually gives you correct data.
So, um, so I wanted to also like, like backstage is open source. Um, and I just wanted to like talk a little bit briefly about like, like that because like now we've been talking about is a key part of how you kind of implemented your culture, uh, as a platform product. Um, and it feels like very much like your secret sauce, but then you went ahead and gave it away as open source.
Like, just tell me a little bit about that. Why did that came to be?
[00:32:52] Pia Nilsson: So we're huge open source enthusiasts. Uh, I think that's where it started. Uh, and then [00:33:00] on top of that, uh, we, uh, were having such internal success back in 2018 or something with Backstage that we started thinking about how to protect it as well from, uh, a migration.
And it started to, we were absolutely depending on backstage for all of our internal. Products and all of our engineers. Uh, so it was, um, tier one service we call it, and the business critical. So if it's down, we can't ship, uh, any code at Spotify, which is as tier one as it gets, basically. So from that standpoint, we started to think about how to protect it, and we, uh, we figured, well, why, why wouldn't, why?
Uh, why not go the Kubernetes path?
[00:33:48] Per Krogslund: Yeah.
[00:33:48] Pia Nilsson: Because we had ourselves just, uh, migrated from our also open sourced orchestration tool, Helio to Kubernetes. They were actually launched on the same day. [00:34:00]
[00:34:00] Per Krogslund: Yeah.
[00:34:02] Pia Nilsson: And then the Kubernetes took off and we had to migrate to it. Uh, of course we did. So, uh, um. That was our own, uh, infrastructural team's decision to do so because it made, made sense, but it was super painful.
Hmm. And we didn't wanna repeat that experience. So, uh, we figured, uh, we have a, a great, uh, track to follow here. We should just do what Kubernetes did and make backstage the number one IDP in the world. This should be the standard. And, uh, and we'll see where, where that takes us. So, yeah.
[00:34:38] Per Krogslund: Uh,
[00:34:39] Pia Nilsson: I think that was sort of the, the starting point of it.
[00:34:42] Per Krogslund: Okay. And then you, um, and you also, you just received a, an award for this the other day, so I think you're, you're on the right track here, right? Mm-hmm. So congratulations to that. So thank you. Um, but now it's also commercial venture. It's not just a open source one. And so like, did you know from the start you want to be commercial or was more like, [00:35:00] let's open source and see what happens, and then that came later.
[00:35:04] Pia Nilsson: Uh, I kind of wanna say a little bit of both. Yeah. I think at the time we were, we, we, we definitely aimed for like, this is, we needed to be the number one IDP or developed portal in the world.
[00:35:16] Per Krogslund: Mm-hmm.
[00:35:17] Pia Nilsson: And so commercialization wasn't far away, but, uh, we, like, we focused on massive scale and adoption in, in adoption in the beginning because that was a real business target for us.
Then sort of outta that grew, people were of course coming to us like, please, can you maintain this for us? Please? Can you sell us your 200 plus plugins because we don't wanna build them. We can pay. Just start selling them, please. So that's sort of came pretty quickly. Um, as the next step.
[00:35:55] Per Krogslund: Yeah. So you just get very clear signals from people who would like to pay you for this and then that, that yes.
[00:36:00] Became like a natural thing. And so, uh, maybe something that's not so natural thing is like a platform team all of a sudden becoming like a, a commercial venture. Um, what was, um, I'm guessing there's been like some challenges in getting to that point of like thinking commercially instead of thinking as a platform team.
[00:36:17] Pia Nilsson: Yeah, it has been. It has been. And maybe surprisingly so.
[00:36:23] Per Krogslund: Yeah,
[00:36:24] Pia Nilsson: because the platform teams are, uh, often super, sort of passionate and proud, uh, to ship the best of breed they can. They can, the best they can ever do, which is best of breed building on the giants, all of this. And still there was still a like a, a bit of a challenge sometimes to sort of transform a team to look at Yes.
You are dogfooding your product.
[00:36:52] Per Krogslund: Yeah.
[00:36:52] Pia Nilsson: On Spotify and learning about the problem with Spotify, but that next step then [00:37:00] is to sell it to the world.
[00:37:01] Per Krogslund: Yeah.
[00:37:02] Pia Nilsson: It. I, I find it so inspiring and fun and, and such a healthy way to build products. Yeah. I can't believe it almost we're so lucky, but it, I guess it is, um, challenging for people, uh, to have several key stakeholders to juggle at the same time.
So of course it adds complexity in, uh, when you have the internal customer, Spotify and 50. Commercial customers are all asking for a bunch of different things. There is this, uh, natural, um, increase of, uh, complexity for primarily the product manager of the domain. Of course, what to prioritize?
[00:37:50] Per Krogslund: Yeah. So all of a sudden, just instead of just.
Prioritizing internals, Spotify needs. All of a sudden you have, yeah, you have a bunch of customers who might, is it, is it, is it very different what customers [00:38:00] want versus what you want? I guess like you as a company is, is like a lot more mature than a lot of these companies and and so on, but like, is it, is it that different, like what customers want compared to what you need?
[00:38:11] Pia Nilsson: Really not. Okay. It's more about the sequencing as exactly what you're saying. Yeah. Um, there are some, some few occasion, uh, uh, scenarios of course where, or examples, I mean, where customers have been asking for some certain sort, sort of skill exchange features that we ourselves never needed. Uh, so we built them for them.
Uh. That's sort of the rare case.
[00:38:36] Per Krogslund: Okay.
[00:38:36] Pia Nilsson: The more common case is we have this huge problem internally because we're leveraging at scale and we have all of these AI tools around us, and we are basically leveraging everything there is to leverage in the world. We, we, we have it. Yeah. Uh, so we are running into these problems pretty early and we need to solve for them inside of the platform mission.
To [00:39:00] for Spotify's sake. And then we kind of need to explain this problem for the customers that yes, you are gonna run into this eventually, uh, had you gone down our path to try it all, this is what we recommend. Hmm. So sort of that's the path I'm seeing as, uh, tread most often actually.
[00:39:23] Per Krogslund: Okay. Alright. Um, so.
Looking a little bit ahead, uh, or I guess it's not even, it is, it is the presence, like talking about AI and, and, and backstage and this Spotify model and so on. It's a, like, AI has been a, a big change for our companies and they see like things rapidly changing. Some developers feel like their field is like entirely new.
It's not what they remember. They remember just. Being able to write code and so on. But as we, we already talked about a little bit, like a lot of your, uh, like half of your PRS is already managed by fleet management. You have a high degree of, of, um, automation and you already [00:40:00] started like trialing a lot of different AI agents and tools and so on.
Um, so, so, um. And also I read the other day that you are like very far actually on what you feel like AI adoption, I guess in, in the grand scheme of things. Maybe we are only just getting started, but like your organization already like is, is jumping right in and you feel like you can do that confidently, um, with the systems and so on that you have in place.
Um, so I feel like, um, you already touched on this a little bit, but what I see. From AI adoption across a lot of companies that they jump head in into the tool itself. Um, and they, they might not have the, the maturity as a company, especially around data and so on, that really requires like a lot of, you can say governance and, and rules and so on and processes, um, to really take full advantage of ai.
So. Again, you're very, I think you're very biased. You have a clear opinion here, but does like having a, like a platform like you do and the access to data and monitoring and so on, enable AI in [00:41:00] these engineering organizations?
[00:41:02] Pia Nilsson: It, it certainly does. Yeah. This is, this is super cool and this is where I think engineering is really heading.
Yeah, I mean we are, we are certainly there and needs to have, uh, like move further of course along that path, but that's exactly where engineering is heading. So what we are talking about internally, for example, uh, at Spotify is engineers are, are, uh, from sort of managing components, they're managing fleets.
Yeah. And for managing fleets, they're orchestrating fleets. So we are be becoming orchestrators of the fleet. That is what sort of the engineering, uh, work that we see for ourselves. Uh, that's how it's changing. Uh, in order to leverage the ai, uh, coding agents, for example. We use a lot of claw then we, you, uh, trialing on Codex and we are using cursor of course.
And, and, uh, whatnot. [00:42:00] So in order to leverage them safely and securely and effectively
[00:42:06] Per Krogslund: mm-hmm.
[00:42:07] Pia Nilsson: We need them to adhere to standards. So one of the things we're looking into a lot right now is how do we supply the rules and, uh, skills as the, um, industries now aligning on skills for the how to actually do a solar problem.
[00:42:26] Per Krogslund: Yeah.
[00:42:26] Pia Nilsson: How do we, um. Distribute that to the local development, um, environment. So that, that is the problem. Sort of where we are at right now, like. We have it centralized already. Uh, so that is the first step, of course, to centralize all your cps, centralize all of your rules, then centralize all of your skills and have them sort of being, build, build a system so that you can add rules, skills, mcps, and they are, uh, aggregated into your component library.
Um. At the [00:43:00] fly on. On the fly. So the second problem is how do we then push that out to the developer, uh, who is working locally, of course, because that's the most effective way to work. So sort of, those are super interesting, I think, uh, challenges that we, we are looking at right now. And for, for that, you need the platform centralization and you need the empowered decentralization as well.
[00:43:26] Per Krogslund: Yeah. And so also I think when you, when we're talking about especially coding agents is the thing right now, because that's the use case, we really determined. It's like that's, that's effective, that works. So production of code has really gone down. It's, it's much faster to produce the code. But then what a lot of companies are seeing is that they have bottle xray that down the system of just like validating, reviewing.
Observability and making sure that they actually deploy something that's solid. Um, is this a pain you feel as well or is this also your system, like kind of remediating that pain?
[00:43:57] Pia Nilsson: Oh, we feel it.[00:44:00]
Well, well, coding agents are great and everyone loves them, but it's, it's just pushing the bottleneck further down. Yeah. And the whole industry knows and sees this. I think we certainly feel it as well. And right now the bottleneck is of course the PR reviews or the review of whatever code, practice you use.
So, uh, we are using prs, uh, and, uh, we have a review, uh, bottleneck. So, and there of course there are a bunch of AI tools that we can leverage. So we're, uh, trialing three, three of them out right now.
[00:44:36] Per Krogslund: Okay.
[00:44:36] Pia Nilsson: As well as building on open source ourselves to try to see sort of which one, uh, of these three approaches actually is most beneficial for Spotify.
Yeah. Uh, so yeah, it, it's super cool. So the bottleneck just keeps, um, uh, being pushed. Okay. And I think, I mean, it goes without saying, but of course the obvious next bottleneck [00:45:00] for the entire world is in operations.
[00:45:02] Per Krogslund: Yeah. Yeah. Is this something you're seeing already or is the, is the PR bottle like kind of keeping that at base or you kind of have a natural slowdown in your prs and then you're not really feeling that on the ops level yet?
[00:45:14] Pia Nilsson: We're not feeling the bottleneck in, um, like, uh, vulnerabilities, uh, et cetera, coming out to the uh
[00:45:21] Per Krogslund: Okay.
[00:45:21] Pia Nilsson: Collection. Yeah. As in the scale of them? Yeah. Uh, no, because the, the bottleneck takes care of that. And the bottleneck is review right now.
[00:45:29] Per Krogslund: Yeah.
[00:45:30] Pia Nilsson: Um, but then we are, we are already in, in production. We're of course, just like many companies, uh, are seeing.
Uh, the increase of, uh, possibly v vulnerable code entering into our production environment. Yeah. Now we have guard guardrails for that. Thank God we have standards, but we do so, but we do see an uptick in, in us catching these things. Uh, more of them now. Yeah. With all of this, uh, this, uh, cursor and uh, CLO [00:46:00] code generated.
Prs.
[00:46:02] Per Krogslund: Okay. Oh, so, so there's actually like a, a decrease in quality from these coding agents that then affects the system further down? Like they introduce more vulnerabilities? Yeah,
[00:46:10] Pia Nilsson: like for sure. They, they, they could build, uh, without the, using the standards, uh, bypass a few things. Uh, but they, we do catch them before they go to production.
[00:46:20] Per Krogslund: Okay.
[00:46:21] Pia Nilsson: Which is super cool. And that's of course, thanks to our standards again, back to, yeah. The basics, uh, what helps human Yeah. Helps agents as well. Yeah. So, uh, but there is an uptick in, of course, more code, more incidents in a sense. Yeah,
[00:46:37] Per Krogslund: of course.
[00:46:38] Pia Nilsson: So I think one of the things also for the industry that we're definitely gonna talk about more is, well, testing becomes absolutely critical, right?
[00:46:48] Per Krogslund: Mm-hmm.
[00:46:50] Pia Nilsson: Because in order to solve the PR. Review bottleneck. You wanna a, you want the bot to review itself?
[00:46:58] Per Krogslund: Yeah.
[00:46:59] Pia Nilsson: How is it [00:47:00] gonna do that? Well, of course it's gonna use tests and, uh. Some engineers don't write tests,
[00:47:10] Per Krogslund: most dislike writing tests. That very true. Yes. Yeah. So yeah, there might be a future where we, uh, we get tests from the AI as well, um, to validate itself or maybe it's just easier ways of writing tests.
Um, yeah. Um, so. I'm just looking at the clock and I think we are nearly running out of time. Um, so I just wanna bring all this together also with respect for your, your schedule and so on. So the Spotify culture and your like internal platform work. We have the AI component in it as well, and like. So backstage, like from the outside part is really the distillation of like Spotify's greatest hits of developer experience so far.
And I'm sure you have much more coming. Um, but looking a little bit ahead, like what are you, what are you most excited about bringing from the insider Spotify out to like a wider audience, [00:48:00] um, in the future?
[00:48:02] Pia Nilsson: Well, I'm most excited about solving platform engineering problems together with our customers or for our customers because they are the same.
It, it's. It's super, super cool to unite sort of all of our platform platform organizations over the world, sort of. So I'm really excited about that. And the, the thing that I'm, uh, really looking at right now together with the team is shipping our fleet management, uh, AI powered faith management, I should say.
Yeah. So actually leveraging whichever code agent you may have already, just like we are. Through the fleet management orchestration system, safely deploy changes across your entire fleet. So we we're shipping that in our portal, which is the backstage SaaS version. Okay. And I think that's gonna really matter to customers.
[00:48:57] Per Krogslund: Yeah. I, I've seen [00:49:00] seeing it from, at the time, seeing it from the inside and without all the AI things in there. I, I think I was, I was deeply en I think I, I just came from Lander at that time, dealing with Log for Jay, and I think when I heard, oh yeah, log for Jay, we dealt with that like in a day and I just spent six weeks at Solando trying to make sense of it.
So, um, that's of course just from the, from the specific security perspective, but I think that's, um, it's a very impressive feat of. Of, of just scaling engineering and removing to, so I'm, I, I can understand why you're excited about that. So I think, um, to wrap up, um, anything you wanna recommend, um, our dear listeners too, and we'll make sure to put all the links and so on.
This, your notes. Any, anything you wanna recommend?
[00:49:38] Pia Nilsson: Oh, I'd love to. Uh, so, uh, please have a, have a listen to the nerd out. Uh, episode that, uh, that's a podcast from, uh, Spotify Engineering Leadership. So Theolo, we call him, uh, Dave Soki, uh, and you know him so well.
[00:49:55] Per Krogslund: Yep.
[00:49:55] Pia Nilsson: Uh, is the host. He's great and he's always, always having some really [00:50:00] cool, uh, people on.
Uh, that he's interviewing. And the, the one that I really wanna push for is the, I think the latest one, which is called, uh, something Ship Happens or something as well. Yeah. Good name. Uh, but this is, this is one, one episode only, and it's about, um. AI fleet management. Yeah. Uh, that we are, we're shipping. So that's gonna be pretty cool for people to hear what we are doing internally.
And the other thing I think really that is powerful inside of Spotify that can help customers also think through AI is, uh, Steve YIs many, many posts.
[00:50:37] Per Krogslund: Mm-hmm.
[00:50:38] Pia Nilsson: Uh, I, I think just reading, reading his, um. Uh, sort of levels of maturity for AI is helpful. Uh, and thinking about, uh, sort of the eight versions of like, how many clouds can you run simultaneously slotting yourself into which one, which one, 1:00 AM I, am I a seven or a, [00:51:00] or, uh, on this, uh, uh, how many kilowatts I can run?
So the, uh. This, I think he has been really inspiring for us because he's really pushing the envelope on how to think about ai. Uh, and also he has some faces going on, so please, uh, that I think that makes sense to, uh, read up on. Last one I wanted to really push for is, um, Brene Brown. Uh, I think, uh, she keeps being very relevant for our industry.
Mm-hmm.
[00:51:32] Per Krogslund: Yeah.
[00:51:32] Pia Nilsson: And, uh, now more than ever with AI we're, we need to think about the people in order to not, uh, go back into some kind of place where we, we let fear rule instead we should sort of welcome change. And she's, uh, doing that in a very grounded, not superficial way, uh, which I really respect. So I think I wanna just push a little for her Strong ground podcast.
Yeah, [00:52:00] uh, whatever book from her you can read, uh, it's gonna be bring, bring Value to you. I believe
[00:52:07] Per Krogslund: that's an amazing set of recommendations and also a good diverse set of recommendations as well. So I think something in every kind of medium. Um, thank you so much for, for this conversation, Pia. I really enjoyed talking to you and so.
This is, this is the wrap for today, and I think, um, as an outro, as always, this show is brought to you by Docker, the open platform for building, shipping and running apps. So thank you for tuning in, and we will see you next time.