Table of Contents
For over 14 years, Gustav Söderström has been at the heart of Spotify's product evolution, navigating the company from a Swedish startup to a global audio giant. As co-president, chief product officer, and chief technology officer, he has a unique vantage point on the shifts shaping not just Spotify, but the entire digital landscape. He sees the internet evolving in distinct eras: from curation, where users organized content, to recommendation, where algorithms took over. Now, we are entering a new, seismic shift—the era of generation, powered by AI. This transition, he argues, will be as transformative as the last, forcing companies to fundamentally rethink product, user experience, and even business models.
Key Takeaways
- The Three Eras of the Internet: The internet has evolved from Curation (users digitize and organize) to Recommendation (algorithms suggest). We are now in the third era: Generation (AI creates).
- Design for AI's Performance: When building with AI, create "fault-tolerant" user interfaces. If an algorithm is only right one in five times, the UI should present five options, not one, to match its performance.
- Move Autonomy Up the Chain: Spotify evolved from highly autonomous "squads" to a more centralized model. Autonomy now sits at the VP level, balancing speed with strategic alignment and avoiding the chaos of thousands of individual strategies.
- Redesigns are Painful but Necessary: Changing a core user experience will always generate pushback. The key is to distinguish between user frustration from broken habits and feedback indicating a fundamentally wrong decision, using both qualitative and quantitative data.
- Embrace Socratic Debate: Leaders owe their teams clear explanations for decisions. The process of articulating the "why" not only builds understanding but also forces the leader to clarify their own thinking.
The Internet's Next Frontier: From Curation to Generation
To understand where product is going, it helps to understand where it's been. Söderström outlines a clear, three-stage evolution of the internet. It began with a simple premise: take physical goods like books or music, digitize them, and let users organize the content. This was the era of curation.
"The internet sort of started with curation," Söderström explains. "You took something... digitized it, and you put it online and then you asked users to curate it. That was a Facebook, Spotify, and so forth."
Soon, a new force emerged. Algorithms became sophisticated enough to take over the heavy lifting of discovery. Instead of relying solely on user-created playlists or friend activity, platforms began to predict what you'd want next. This was the era of recommendation, a massive shift that required companies like Spotify to "rethink the entire user experience and sometimes the business model as well."
Today, we stand at the threshold of the third major wave: generation.
I think what we're entering now is we're going from your curation to recommendation to generation, and I suspect it will be as big of a shift.
This new era isn't just about better recommendations; it's about creating entirely new content and experiences. From text to images and now music, generative AI makes it possible to build products that were previously science fiction. For product leaders, this isn't an incremental change—it's a call to once again rethink everything.
Building for the Generative Era: Lessons from Spotify's AI DJ
Spotify's first major foray into a purely generative product is the AI DJ, an experience that simply "couldn't have existed without generative AI." The feature uses a synthesized voice to create a personalized radio host that introduces songs and artists based on your listening habits. The project provided several core principles for building in the generative era.
Design a Fault-Tolerant User Interface
A common mistake in the early days of machine learning was designing overly simplistic interfaces that didn't account for the algorithm's error rate. A single, giant "play" button only works if your prediction is 100% accurate, which it never is. The key, Söderström notes, is to match the UI to the performance of the underlying model.
He points to Midjourney's interface as a prime example. "What they did was they generated four simultaneous low-res images very quickly... apparently their performance was probably one in four," he says. This "fault-tolerant" design gives the user a high chance of finding one good result immediately, building trust and engagement. The UI must provide an escape hatch, allowing users to easily correct the algorithm when it’s wrong.
Get Out of the Way
With powerful new technology, the temptation is to show it off. The AI DJ team, however, established a crucial principle: do as little as possible and get out of the way. Users came for the music, not to hear a synthesized voice talk endlessly. "It is trying to get you to the music," Söderström emphasizes. "And I think that's why it's working." The technology serves the core user need rather than becoming the center of attention.
AI as a New Instrument
The rise of generative music has sparked debate about authenticity and artists' rights. Söderström draws a parallel to the arrival of digital audio workstations, which enabled artists like Avicii—who couldn't play traditional instruments—to create globally acclaimed music. Initially dismissed by some as "not real music," it's now a celebrated genre.
He views generative AI not as a replacement for artists, but as a new, more powerful instrument. "Being original is very hard," he notes. As AI makes it easier to create generic music, the value of true, human-driven uniqueness will only increase. The challenge now, as it was with the shift from piracy to streaming, is to build the business models that allow creators to benefit from this new technology.
Rethinking Team Structure: Why Spotify Moved Beyond Squads
Spotify became famous in tech circles for its "squads, tribes, chapters, and guilds" model, which emphasized small, autonomous, full-stack teams. While effective in the company's early days, this structure created challenges at scale. Spotify has since evolved, moving away from the very model it helped popularize.
The Problem with Hyper-Autonomy
The original squad model, inspired by Sweden's bottom-up culture, granted extreme autonomy to teams at the "leaves" of the organization. While this empowers smart people, it can also lead to inefficiency. "You're going to have 100 squads with 100 strategies running in 100 directions," Söderström says. This can create a fragmented user experience and a lot of wasted effort, or "heat."
Finding the Right Altitude for Autonomy
The solution wasn't to centralize all decision-making at the top, which creates bottlenecks. Instead, Spotify found a middle ground, placing the primary locus of autonomy at the Vice President level. This structure creates a cohort of senior leaders—dozens or hundreds, not thousands—who have the context, experience, and strategic alignment to make big bets. It allows for diverse thinking without devolving into chaos.
This organizational design reflects a strategic choice. Söderström contrasts two successful extremes: Amazon's decentralized, competitive "two-pizza teams" and Apple's highly centralized, functional organization. Because Spotify's strategy is to deliver a single, coherent user experience across music, podcasts, and audiobooks, it has chosen to lean more toward Apple's centralized model. "We think the user experience in keeping that simple is the most important thing," he states.
The Science of Big Bets: Navigating Redesigns and User Feedback
In a recent launch event, Spotify unveiled a major redesign of its home feed, shifting to a vertical, discovery-oriented format reminiscent of TikTok. The user reaction was swift and polarized. For Söderström, this experience reinforced critical lessons about making big bets and interpreting feedback.
The Problem: Breaking the Taste Bubble
The goal of the redesign was to solve a common user complaint: getting trapped in a "taste bubble." Spotify excels at recommending the next song within a genre you already love, but it struggles to introduce you to something entirely new. Why? Because the "hit rate" for novel recommendations is inherently low. Inserting a random reggaeton track into a metal playlist feels broken.
A feed format, where users can quickly swipe through content, is better suited for low-hit-rate discovery. "You need something where one out of ten is a success," Söderström explains. "If you find one gem out of ten tries, you're very happy."
The Learning: Recall vs. Discovery
The team put the new discovery-focused feeds on the home screen. While the feeds worked well for their intended purpose, the change disrupted a fundamental behavior. Quantitative data revealed that users engage in "recall"—navigating to a known playlist, podcast, or session—on the homepage nearly 90% of the time. The redesign had flipped that ratio, prioritizing discovery over recall.
The negative feedback wasn't about the discovery features themselves. "They're complaining about the things they don't get anymore," he clarifies. Users couldn't find their favorite playlists. Söderström compares it to someone rearranging your physical desk; even if the new layout is theoretically more efficient, your muscle memory is broken, causing immediate frustration.
Strong Opinions, Loosely Held
Navigating this feedback requires separating frustration with change from a genuine design flaw. The key is a scientific approach: analyze quantitative data (like users shifting to search to find their library), conduct user research, and be willing to change your mind.
You have to believe in things 100 percent until the data says no, and then you believe in something else 100 percent.
This philosophy, often called "strong opinions, loosely held," is difficult but essential. The team learned they had underestimated how well the original homepage supported recall. The new hypothesis is to make the discovery tools prominent and accessible but voluntary, preserving the efficient recall functionality that users depend on.
Leadership, Clarity, and Peeing in Your Pants
Söderström's approach to product leadership is grounded in first-principles thinking, clear communication, and a few memorable analogies.
The Power of Explanation
He fosters a culture of "Socratic debate," where the best idea wins, regardless of seniority. A core tenet of this is that leaders must be able to explain their reasoning. "You owe an explanation," he insists. This isn't just about transparency; it's a tool for better thinking.
The best way to understand something is to try to explain it to someone else.
If you can't articulate your logic, you probably haven't fully understood the problem yourself. He pushes his teams to treat product development as "100 percent science," forcing a rigor that surfaces flawed assumptions and strengthens strategy.
The Pee-in-the-Pants Analogy for Short-Term Thinking
To caution against short-term decisions that create long-term problems, Söderström uses a vivid, if unconventional, Swedish analogy. Making a short-sighted choice is like "peeing in your pants in cold weather." He explains: "It feels really warm and nice to begin with, and then after a while you start to regret it." The memorable phrase serves as a quick, effective shorthand for prioritizing sustainable, long-term thinking over immediate gratification.
Conclusion
From navigating technological shifts to restructuring a global organization and weathering the storm of a public redesign, Gustav Söderström’s journey offers a masterclass in modern product leadership. His approach is rooted in a deep curiosity about how things work, a willingness to challenge long-held assumptions, and a commitment to rigorous, scientific thinking. As the internet enters the generative era, product leaders will need more than just intuition. They will need clear frameworks, the courage to take big swings, and the humility to listen to the data when it proves them wrong. And perhaps, a memorable analogy or two to guide the way.
