Inside Spotify’s Machine: Innovation, AI, and Building in Sync
Podcast Review: “How Spotify Thinks” — Invest Like the Best, with Gustav Söderström
What does it take to steer a 700-million-user platform through a generational shift in technology?
Gustav Söderström is the Co-President, Chief Product Officer, and Chief Technology Officer of Spotify—and the architect behind how the company thinks, builds, and bets.
In this conversation with Invest Like the Best, Gustav opens up about the systems behind Spotify’s innovation engine: how they prioritize bets, synchronize global teams, and prepare for an AI-native future.
From restructuring their backend to make AI usable, to designing a product culture that embraces failure and demands real explanations—this episode is a masterclass in modern product leadership.
1. AI or Die – A New Macrowave Is Here
The episode opens with an idea that should stop any founder in their tracks: AI or Die. Gustav Söderström, Spotify’s Co-President and product chief, sees artificial intelligence not as a feature or an add-on—but as a macrowave shift that no company can afford to ignore.
He compares it to previous tech inflection points:
“Just like it was ‘Smartphone or Die’, before that ‘Internet or Die’, and before that ‘Computer or Die’. This is one of those shifts—it’s not your choice whether you adopt it or not. It’s going to happen to you.”
That framing matters. Spotify’s strategy isn’t just to adopt AI—it’s to reposition the entire company to ride the wave. Gustav calls this internal philosophy “surfing the macro wind,” and it’s more than metaphor. In prior tech shifts—like the smartphone era—Spotify had to rebuild its business model from the ground up. Originally, their free tier was desktop-only. But as mobile became the default device, millions of new users had no desktop access at all. Spotify had to rethink how to offer free mobile listening without cannibalizing premium.
This is the lens they’re using for AI today: not just what features should we ship?, but does this force a business model change?
This distinction—product vs. model—is critical for founders. Many companies will bolt on AI. Few will change how they monetize, serve users, and structure teams. Gustav is clear: AI is horizontal. It will touch product, productivity, organizational design, and cost structure. If you treat it as a “tools upgrade,” you’ll get left behind.
2. The Uplink Revolution – Language as Input
Spotify’s most radical observation is deceptively simple: most apps are dumb on input.
Think about your phone. You stream videos, you scroll feeds—tons of “downlink.” But your “uplink”—what the app hears from you—is mostly just taps and skips. Gustav calls this asymmetric bandwidth. Spotify, like most consumer apps, has rich information flowing to the user… and nearly nothing coming back.
Enter generative AI.
“What changes in the age of generative AI is that the uplink can now be English language. It can be almost as rich as the downlink.”
This one shift flips the architecture of product design. Spotify’s first real-world application of this idea is AI playlisting—now live in 40+ countries. Users type what they want in plain English:
“Give me a high-energy running playlist. EDM only. 160 BPM. No vocals. Big drops.”
The system uses an LLM trained on the user’s history, public knowledge, and catalog metadata to generate playlists that reflect exactly what’s in the user’s mind.
This is huge. Before, Spotify had to infer taste based on skips, likes, or playlist behavior—imperfect proxies. Now, users can just tell the system what they want.
And it’s not one-off. Gustav sees this as a foundational shift:
“Consumer products are going to be a conversation… not a service you use.”
Founders should take note. If your product still relies on clicks or static filters, you're not just behind—you may be fundamentally misreading where interface design is going.
In a generative world, products don’t wait for users to learn them. They learn the user.
3. Inside the Innovation Engine – How Spotify Picks Bets
Spotify doesn’t innovate by chance. It innovates by system.
At the heart of Gustav Söderström’s approach is a structured process he calls the “bets board.” It happens every six months. Here’s how it works:
All of Spotify’s VPs—about 14 of them—pitch their top initiatives as if they were startups. They don’t appeal to executive preferences or legacy priorities. They’re pitching into a stack-ranked system. Gustav and his co-president Alex Nordström then rank all the bets globally, one to forty (or fifty), based purely on strategic alignment and value.
Yes, rank, not cluster, not prioritize—stack rank.
This clarity eliminates internal friction. Teams don’t fight for budget or make backroom deals. They see the ranked list. If your bet is #12 and someone else’s is #5, the organization executes from the top down until it runs out of capacity.
And here’s the genius: the process forces hard decisions up front.
Gustav points out that when leaders say two things are equally important, they’re just kicking the decision downstream. That’s when middle managers start to fight over headcount and timelines. Instead, stack ranking acts as a forcing function: “If you had to kill one of these two bets, which would it be?” That’s strategy in action.
Spotify complements this with a prototype-first culture. Before any six-month plan is locked in, teams build high-fidelity mockups of what Spotify could look like if certain bets are implemented. This creates alignment early, and prevents nasty surprises halfway through development.
It’s a framework any founder can adapt:
Six-month planning cycles.
VP-level pitching.
Stack ranking, not committees.
Resource from the top down.
Visualize before you build.
Bottom line: Spotify treats product management like portfolio management. Structured, probabilistic, high-leverage.
4. The E-Team – How Real-Time Alignment Happens
Here’s where the Spotify machine really shines.
Every Tuesday for three hours, Gustav, Alex, and the company’s VPs meet in what they call the E-Team (Execution Team). It’s not a planning call. It’s a real-time operating room. If someone is blocked, if two teams are misaligned, or if execution is off track, it gets addressed in the room.
One rule?
“You’re not allowed to say ‘let’s take this offline’—because everyone you need is right there.”
This simple but strict discipline reduces drag across the entire organization. Gustav says that because of this rhythm, no team is ever blocked for more than 2.5 working days. That’s the maximum time before another Tuesday E-Team session occurs.
Another subtle but brilliant rule: no direct reports allowed.
Why? Two reasons:
It forces VPs to deeply understand their domains. No delegating updates.
It builds strong interpersonal trust among senior leaders over time.
Without rotating attendees or proxy presenters, the team becomes tight-knit, aligned, and accountable.
This level of cross-functional fluency is rare. Gustav explains that his product and engineering leads understand Spotify’s P&L and gross margin targets. And on the flip side, his business leads know what a monorepo is, and how tech debt impacts velocity.
This fluency doesn’t just keep people informed—it keeps Spotify synchronized. When they ship a new app version or UI change, every vertical—music, podcast, books, personalization, platform—is aligned.
To founders: this is what operating rigor looks like at scale. No misalignment theater. No “weekly syncs” that hide decisions. Spotify’s system works because it resolves tension quickly, visibly, and with the right people in the room.
5. AI’s Real Bottleneck – The Backend, Not the Model
Everyone wants to talk about LLMs and prompt engineering. But at Spotify, the hardest part of scaling AI isn’t the model—it’s the backend.
Gustav calls it out bluntly: the real work is “old-school engineering.” Before Spotify’s teams can use AI to reason over 15 years of user data—play history, taste graphs, family plans—they have to expose that data in real-time APIs.
Here’s the context: In the era of traditional machine learning, historical data was often stored in cold storage or pulled in massive offline jobs. It was optimized for training, not for querying.
But generative AI requires a different setup. You can’t generate dynamic reasoning or build interactive systems like AI playlisting or real-time user modeling unless your systems are accessible, live, and interoperable.
So what’s the unlock?
Spotify is adopting something called the **Model Context Protocol (MCP)**—a standard that wraps internal services so they can be queried in natural language. That means PMs, designers, and even business leads can prototype using tools like Cursor without writing production code.
Example:
A PM screenshots a screen in Figma, uploads it to Cursor, and says:
“Wire this up with the user’s liked songs and preview 30 seconds of each.”
If services are wrapped in MCP, the system can generate a live prototype by reasoning across the product stack. This is how AI goes from code assist to company-wide interface.
Takeaway: If you’re building a product organization and thinking “AI isn’t helping much yet,” ask yourself—can your infrastructure even talk back?
Spotify’s not betting on faster models. It’s rebuilding its internals to let AI become an orchestration layer, not just a clever chatbot.
6. Product Overhang – The Gold We’re Sitting On
One of the boldest insights in the podcast is this: the most powerful AI use cases haven’t even been built yet. And not because of model limitations—but because of organizational underreach.
Gustav calls this the “product overhang.”
His analogy: When spreadsheets arrived, people feared accountants would be obsolete. Instead, we got more accounting. The cost of modeling dropped, so companies built dozens of scenarios they never would’ve attempted before. The same is true of inference.
“We’re going to be ashamed of how mundane the things were that we used inference for.”
Gustav imagines a near-future where you use reasoning engines to answer everything:
“Can we launch music videos in Brazil next quarter?”
“What’s our licensing risk if we add this feature?”
“How has my music taste changed in the last year?”
These questions require synthesis, not just retrieval. That’s where current AI can shine—if we build the scaffolding to support it.
And that’s the tension: we already have models good enough to unlock new categories of product, but companies are treating them like novelties. They’re stuck in prompt hacks, one-off summaries, and low-leverage applications.
Spotify’s leadership doesn’t just see this as an opportunity—they see it as a strategic gap. Startups are moving faster because they don’t have 15 years of cold storage to restructure. But that won’t last forever.
If Spotify succeeds in bringing its infrastructure online, it could set the benchmark for what an AI-native org looks like at global scale.
To founders: if you’re not building for where the models are headed, you’re building on top of a shrinking opportunity. AI is not a feature race—it’s a product architecture race.
7. Failing Forward – The Story of “Moments” and a Culture of Risk
It’s rare to hear a senior executive admit to a product failure this openly. But Gustav doesn’t just share his missteps—he explains why they mattered.
Years before TikTok popularized swipe-based media feeds, Spotify attempted something similar. Gustav spearheaded an ambitious redesign called Moments—a UI where the app would immediately start playing music, and users would swipe between moods and genres. No search, no clicks. Just sound your way to what you wanted.
There were two major problems:
The tech wasn’t ready — Spotify’s machine learning systems weren’t mature enough to power this kind of real-time taste navigation.
The UX was polarizing — users didn’t like music autoplaying without consent. Many found it jarring.
Despite internal skepticism, the team pushed hard. They even announced the product publicly, complete with launch videos. But when they rolled it out, the results were underwhelming. Worse: they discovered a bug in the A/B testing system, and the real performance numbers were far worse than expected.
The feature was scrapped. A year of work was lost.
And Gustav? He didn’t get fired. Daniel Ek, Spotify’s CEO, responded with a principle borrowed from Jeff Bezos:
“I judged you by the inputs, not the outputs.”
This mindset—focusing on reasoning and process rather than outcomes—enabled Gustav to take bigger bets going forward. The experience became part of Spotify’s DNA: launch with structure, test with care, and learn in public.
To founders, the lesson is powerful: the real cost of failure isn’t the feature that flopped—it’s the cultural damage if your team fears trying again. Spotify made failure survivable, and as a result, kept its best talent bold.
8. The Ultimate Bundle – Spotify’s Subscription Strategy
Spotify’s product may be beautifully simple on the front-end—but behind it is one of the most complex monetization engines in the consumer web.
Gustav is clear: the product strategy is driven by a single app philosophy. Music, podcasts, audiobooks—everything flows through one install, one experience. That’s not just UI discipline. It’s a distribution advantage.
Why?
App install friction is real. On iOS and Android, the average user downloads less than one app per month. So if Spotify wants to expand into new formats (books, video, learning, etc.), it makes more sense to add verticals into the core app than launch separate products.
This leads to what Gustav calls the ultimate bundle.
Rather than differentiating via exclusive content (which Spotify tried briefly in podcasting and abandoned), they now differentiate on:
Format variety (music + podcasts + books)
Product quality (personalization, cross-format discovery)
Pricing power (increasing value-per-dollar for users)
The strategic shift away from exclusivity is worth emphasizing. Spotify learned the hard way: betting on celebrity hosts and exclusive shows created cost bloat without clear user lift.
“They were celebrities, but not always great podcast hosts,” Gustav admits.
Now, they’ve reverted to platform fundamentals—max catalog access, smart distribution, and letting creators go wide.
And it’s working.
More retention: Adding podcasts and audiobooks increases time spent.
More willingness to pay: The bundle makes Spotify a better deal, justifying price increases.
More scale: All formats contribute to engagement, without cannibalizing one another.
Founders should study this carefully. Many products try to expand too late or with the wrong economics. Spotify shows that if you control the front door (the app), you can add rooms (formats) without collapsing the house.
Bundling isn’t just a business model—it’s a retention and differentiation strategy.
9. Looking Ahead – Bundles, Billion-User Ambitions, and Business Model Flexibility
If Gustav Söderström had to sum up Spotify’s next five years, it would sound like this:
“I hope we’ve cracked the billion-user line... and added more verticals to the subscription.”
The vision isn’t just about more users—it’s about more value per user. Spotify sees itself not as a music service, but as a multi-format media subscription platform. The goal: to be the most valuable bundle in a world where consumers increasingly resist fragmented, paywalled content.
But Gustav also hints at something deeper: a new layer of differentiation through bundling strategy.
Where others compete on exclusive content, Spotify competes on:
Seamless cross-format consumption
Predictable pricing for diverse value
A single, unified app experience
That’s not easy. Internally, it means reconciling wildly different business models:
Music = pooled royalties
Podcasts = ad-supported and partner opt-in
Audiobooks = streaming hours included, with optional top-ups
Most companies would split these into separate products. Spotify does the opposite. It manages the internal chaos so users don’t feel it.
This orchestration layer—the “Spotify Machine”—is critical. It handles the logic of personalization, format switching, cost attribution, and UI coherence. It’s also the reason why adding a new vertical doesn’t fragment the product.
For founders, Gustav’s message is clear: if you want to win on bundling, you don’t just need content—you need architecture. Infrastructure. An internal operating system that protects the user from your own complexity.
10. Final Takeaways – Process Over Heroics
There’s no singular “Spotify story” in this episode. Instead, there’s a pattern. Spotify wins not by gambling on genius, but by designing an org that learns faster than its peers.
Here are the lasting lessons:
Frameworks over instincts: Gustav arms his teams with mental models (Seven Powers, Value Stick, Better Simpler Strategy) so they speak a common strategic language.
Explanations over intuition: He draws from David Deutsch and Popper to push for hard-to-vary theories, not gut calls. He even delays product launches that “win” A/B tests—until teams can explain why the variant worked.
“Pattern recognition is useful—that’s seniority. But explanation scales across the org.”
Failure tolerance over risk aversion: The Moments UI story is a case study in leadership maturity. Daniel Ek didn’t punish the failure—he doubled down on the logic that led there.
Synchronized structure over chaotic agility: Weekly 3-hour E-Team meetings, no delegation, no “offline” handoffs. Spotify’s ability to coordinate product, design, data, engineering, and licensing—at global scale—is one of the most underrated stories in tech.
Process over personality: The podcast ends with Gustav’s quiet tribute to Daniel:
“What made me excel was that he gave me room to fail—and made me feel safe to keep taking risks.”
Spotify’s magic isn’t a feature. It’s the machine behind the features. A machine that runs on debate, structure, forgiveness, and iteration.
For any founder building a company in the age of AI, this episode isn’t just a podcast. It’s a case study in how to architect a learning organization—and how to stay ambitious when the road gets hard.