Lenny's PodcastThe science of product, big bets, and how AI is impacting the future of music | Gustav Söderström
CHAPTERS
- 0:00 – 0:45
From curation to recommendation to generation: the AI era shift
Gustav frames the evolution of the internet as three eras: user curation, algorithmic recommendation, and now generative creation. He argues the generative era will be a similarly disruptive shift that will likely require rethinking products and interfaces from first principles.
- •Early internet success came from digitizing content and letting users curate it
- •Recommendation systems replaced human curation and forced UX/business-model rewrites
- •Generative AI represents a new era, not just “more ML”
- •The right UI/UX for the generative era is still unclear
- 0:45 – 4:10
Show setup: why Gustav matters + sponsor interlude
Lenny introduces Gustav’s role at Spotify and previews the themes: big bets, org design, AI in product, and the future of music. A sponsor segment follows before the interview begins.
- •Gustav’s scope: product + technology strategy at Spotify
- •Episode promises: AI impact, org structure, big UI bets, and product principles
- •Sponsor messages before the conversation starts
- 4:10 – 6:55
Gustav’s path to Spotify and the mobile business-model puzzle
Gustav recounts his entrepreneurial and Yahoo background and how he joined Spotify to lead mobile. He explains why mobile forced both product and business model innovation (network constraints and ad-funded streaming limitations).
- •Joined Spotify in 2008/2009 to define the mobile strategy
- •Early mobile couldn’t stream reliably (EDGE era), pushing offline/other approaches
- •Mobile also challenged the ad-funded free model
- •Over time he expanded from mobile → product leadership → CTO responsibilities
- 6:55 – 12:38
Why he launched a podcast: creator empathy, culture, and recruiting leverage
Gustav shares the motivations behind producing a high-production Spotify story podcast: a personal creator drive, deeper empathy for podcast creators, and a scalable way to build internal culture. The internal version proved effective, and the external version became a recruiting and transparency tool.
- •Personal interest in storytelling/writing as a “secret creator dream”
- •Hands-on learning: rights and production challenges (e.g., music in podcasts)
- •Internal podcast as culture-building and leadership accessibility mechanism
- •External podcast as a way to demystify Spotify leadership and attract talent
- 12:38 – 21:24
How product teams should approach AI: new era, new UI rules (AI DJ case study)
Gustav distinguishes recommendation-era ML from generation-era ML and urges teams to treat generative AI as qualitatively different. He walks through Spotify’s AI DJ as an example of a truly generative product and shares principles like fault-tolerant UIs and resisting the urge to “show off the tech.”
- •Generative AI should be treated as a new paradigm, not incremental ML
- •Many near-term uses: better recommendations, safety classification, cultural context
- •AI DJ as a product that couldn’t exist without generative AI (voice + scripting)
- •Principle: design fault-tolerant UIs aligned to model hit-rate + provide escape hatches
- •Principle: minimize narration—get out of the way and deliver the core value (music)
- 21:24 – 26:20
AI-generated music: instruments, originality, and the rights/business-model reset
The conversation turns to generative music and what it means for artists, originality, and the industry. Gustav compares today’s skepticism to earlier transitions (e.g., EDM tools and Avicii), and argues the biggest unresolved piece is a rights and compensation model akin to the post-piracy shift.
- •Generative music likely becomes a new “instrument,” enabling new creator types
- •The AI-vs-real-music distinction breaks down; it becomes a spectrum of AI assistance
- •Models are great at “like existing music,” but true originality remains hard—and valuable
- •Parallel to piracy: technology shift first, then business-model innovation to compensate rights holders
- 26:20 – 28:23
The “magic trick” requirement for great products—and why it fades
Gustav explains his belief that breakout products often feel like a magic trick on first use. He uses AI DJ (and tools like Midjourney/GPT) to show how product iteration and performance thresholds create that moment—and how it inevitably becomes normalized.
- •“Magic” is a useful product outcome signal (often correlates with virality)
- •AI DJ’s initial magic: personalized spoken content at massive scale
- •Crossing the “magic line” often requires careful scoping and fine-tuning
- •Even researchers experience residual ‘magic’ due to incomplete understanding of LLM behavior
- 28:23 – 36:03
Spotify org evolution: beyond squads, redefining autonomy, and VP-level strategy ownership
Gustav addresses Spotify’s famous squads/tribes model and why it doesn’t scale well as-is. He describes shifting to larger, more traditional teams and rethinking where autonomy lives—landing on high autonomy at the VP layer to avoid both “leaf-level chaos” and top-level bottlenecks.
- •Squads were useful early, but scaling in units of ~7 creates overhead at large scale
- •Extreme leaf-level autonomy + junior org can create “100 squads, 100 strategies”
- •Top-only autonomy bottlenecks decision-making and throughput
- •Spotify’s current approach: meaningful autonomy concentrated around VP level
- •Example: VPs define strategies for areas like podcasting, experience, and personalization bets
- 36:03 – 43:34
Centralized vs decentralized orgs: Apple vs Amazon, hard APIs, and “shipping the org chart”
Gustav lays out an organizational spectrum from decentralized speed (Amazon) to centralized coherence (Apple). He explains how Amazon’s internal competition requires strong enforced APIs, and why Spotify leans more centralized to keep one app’s UX coherent across multiple content types and business models.
- •Decentralization speeds time-to-user but risks duplicative UX and fragmented experiences
- •Amazon’s competition incentives require centrally enforced “hard APIs” to cooperate at scale
- •Centralization (Apple-like) improves UX coherence but can reduce shipping speed
- •Spotify chooses centralization due to one unified app spanning music/podcasts/audiobooks
- •Cross-vertical recommendations and unified UI require coordination and shared systems
- 43:34 – 54:08
The Home redesign big bet: solving “taste bubbles” with feeds—and what went wrong
Gustav breaks down the rationale for Spotify’s feed-like discovery experiences: breaking users out of taste bubbles requires a low hit-rate, low-cost interaction model. The misstep was pushing too much discovery onto Home, unintentionally undermining the ‘recall’ use case (quickly returning to known playlists/sessions).
- •Taste-bubble problem is different from normal recommendations (low signal by definition)
- •Discovery needs low switching cost + fast iteration over candidates (feed/swipe paradigm)
- •Sub-feeds (music/podcasts) worked as intended for discovery workflows
- •Putting feed-first discovery on Home flipped behavior from ~90% recall to ~90% discovery
- •User backlash largely reflected recall friction (“I can’t find my stuff anymore”)
- 54:08 – 1:02:35
How they tested, learned, and iterated: separating ‘change pain’ from real product flaws
Gustav explains the unique risk of redesigns: users can’t opt out, so feedback mixes habit disruption with genuine usability regressions. He outlines how Spotify used A/B testing, cohort analysis, quantitative shifts (Home → Search/Library), and user research to refine the hypothesis toward “discovery is available, but voluntary.”
- •Redesigns differ from features: participation is mandatory, so backlash is inevitable
- •Two types of anger: ‘you changed things’ vs ‘you made it worse’—hard to distinguish
- •Signals: traffic migration patterns and misuse of discovery UI for recall needs
- •Methods: A/B variants, user research observation, parsing qualitative feedback carefully
- •Operating principle: strong opinions loosely held—update beliefs quickly when data contradicts
- 1:02:35 – 1:10:07
Operating cadence and leadership craft: 10% planning, clarity through explanation, walk-and-talk thinking
Gustav shares a pragmatic planning heuristic: keep planning to ~10% of the cycle to avoid over-planning and under-executing. He also describes how he drives clarity through Socratic debate and forced explanation, plus a surprisingly effective habit: walking one-on-ones for better thinking and strategy work.
- •“10% planning time” rule of thumb (e.g., 2 weeks planning for a 6-month cycle)
- •Energy comes from working on things he genuinely believes in and finds novel
- •Clarity as a leadership obligation: employees may not agree, but deserve to understand why
- •Explanation as a test of understanding (if you can’t explain it, you may not get it)
- •Distributed “walk-and-talk” one-on-ones improved creativity, especially during the pandemic
- 1:10:07 – 1:24:30
Rapid-fire close: pee-in-your-pants analogy, Sweden in Succession, what’s next for podcasting, lightning round
The conversation lightens with Gustav’s ‘peeing in your pants’ metaphor for short-term decisions, plus his take on how Succession portrays Swedish culture. He then shares directional priorities for Spotify podcasting (discovery and monetization) before finishing with a lightning round on books, products, and rituals.
- •“Peeing in your pants” = short-term comfort that creates long-term regret
- •Succession’s Scandinavian depiction: authentic elements, exaggerated negotiation style
- •Podcasting priorities: creator discovery/audience growth + stronger monetization options
- •Consumer-side improvements: better UX, device ubiquity, and AI-enabled enhancements
- •Lightning round: book recommendations, favorite products (ChatGPT, Duolingo), and “think it/build it/ship it/tweak it”
