Lenny's PodcastThe science of product, big bets, and how AI is impacting the future of music | Gustav Söderström
At a glance
WHAT IT’S REALLY ABOUT
Spotify’s Gustav Söderström on AI, big bets, and product orgs
- Gustav Söderström, Spotify’s co-president, CPO, and CTO, discusses how the internet has evolved from curation to recommendation and now to AI-driven generation, and what that means for product strategy and UX. He shares Spotify’s approach to AI (including the AI DJ), how they structure product and engineering teams after moving beyond the famous “squads” model, and how they think about big, risky bets like their controversial home feed redesign. Gustav also explores the future of AI-generated music, the need for products to feel like “magic,” and the organizational tradeoffs between speed, autonomy, and coherence. Throughout, he emphasizes fault-tolerant interfaces, scientific A/B testing, clear strategy ownership at the VP level, and a culture of explaining decisions with rigor rather than relying on “magic.”
IDEAS WORTH REMEMBERING
5 ideasTreat generative AI as a new paradigm, not just “better recommendations.”
Gustav frames the internet’s evolution as curation → recommendation → generation, arguing that generative AI is qualitatively different from past ML and will eventually force teams to rethink product experiences, interfaces, and even business models from first principles.
Design fault-tolerant UIs that match your model’s real performance.
You must understand hit rates and error patterns to design the right interface—e.g., Midjourney shows 4 low-res images because its success rate might be 1-in-4, and Spotify’s AI DJ includes a clear ‘escape hatch’ for users when the system is wrong.
Aim for “magic trick” moments by scoping and tuning until it feels impossible.
Products that go viral often do one thing so surprisingly well that users can’t see how it’s done (e.g., GPT-4, AI DJ). This typically comes from ruthless scoping and performance tuning rather than piling on features or “showing off” the tech.
Place real autonomy at the VP layer to balance speed and alignment.
Spotify learned that putting autonomy at the squad level produced “100 strategies from 100 squads,” and putting it at the CEO level bottlenecks decisions; instead, VPs (e.g., for podcasts, personalization, UX) own strategy and big bets within a centralized product structure.
Separate discovery from recall and design explicitly for each use case.
Users use Spotify’s home mostly (≈90%) for recall (getting back to known content) rather than discovery; pushing a TikTok-style discovery feed onto home broke recall, teaching Spotify to keep recall dense and predictable while making discovery powerful but optional and well-signposted.
WORDS WORTH SAVING
5 quotesWe’re going from curation to recommendation to generation, and I suspect it will be as big of a shift that you will eventually have to rethink your products.
— Gustav Söderström
Don’t think of this as just more of the same machine learning. The recommendation era was one type of machine learning. The generation era is a different type.
— Gustav Söderström
If you’re renting all of this expensive brainpower and then you give them no room to think for themselves, that doesn’t sound smart.
— Gustav Söderström
You have to believe in things 100% until the data says no, and then you believe in something else 100%. That sounds easy. It’s very hard to do.
— Gustav Söderström
We don’t promise everyone that they have to agree, but even if they don’t agree, they should be entitled to understand why you’re making the decision.
— Gustav Söderström
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome