Spotify Studio Killed the Podcast Star

Welcome to Product Cocktail, where the takes are as polarizing as a shot of Fernet—but the insights come together like a perfectly crafted daiquiri.

In May, Spotify announced Studio by Spotify Labs at Investor Day, which they describe as "a standalone app built for Spotify that understands your taste, recommends music, and creates Personal Podcasts from your inbox, calendar, and much more."

In other words: an AI Agent, Spotify's Version.

I got access to the Research Preview in late-September. I spent a few weeks using it, so you don't have to.

The curious thing: Studio is two distinct products in one.

Door number one: the taste agent. A self-proclaimed "powerful" (lol) agent sitting on top of your compendium of Spotify data. It will wax poetic about your listening story, make you a highly tailored playlist based on your vibe, and even reorganize your library (how Marie Kondo of them). It feels like someone put ChatGPT on top of your Spotify library, and it's kind of cool. Think: AI DJ but it doesn't just take requests, you're sitting in the producer chair guiding the direction of the entire show.

Door number two: a NotebookLM competitor? The Tyler Durden to Spotify's music agent Narrator is an AI podcast generator. Spotify says it will make you "daily briefings" (snore), a "learning journey," or a podcast on a particular topic.

To be clear, Studio isn't (yet?) trying to storm the copyright minefield-laden beaches alongside infamous lawsuit magnet, Suno.

Studio is an OpenClaw-style assistant with a Spotify login. It's the same shape: an agent running on your own computer, wired into email and calendar (although that's less emphasized), messageable from your phone, running scheduled tasks, and needing an always-on computer.

Spotify knows that it's the music data GOAT. That advantage has informed their product development for over a decade — Spotify Wrapped, daylist, AI DJ, etc. The Studio FAQ page answers "What makes this different from other AI agents?" by contrasting other agents, which start from scratch with each user, with Studio, which "starts from years of your actual listening: your artists, your routines, your taste at 2am versus Saturday morning."

Even the copy answering "What can Studio do for you" emphasizes that it can read you like an album liner: "It understands your taste."

Then, they drop this edge like it's hot.

Spotify spends the rest of their landing page yapping about the Personal Podcasts feature they folded into this agent. Daily audios, deep dives, travel guide, oh my.

When you launch the agent, its first words are "What should we do first?" and it answers its own question like a call-and-response song, suggesting these three podcast-coded options.

Even its helpful "Ideas" tab doesn't mention music until you've scrolled past its featured section into "More Ideas."

To wit, my entree into Studio was a Spotify desktop in-app notification, so I didn't even know its taste agent capabilities existed until I started to write this newsletter. (The landing page wasn't part of my onboarding, it was purely ex post facto product research.)

The problem is that the taste agent is better.

For one, the podcast generator is a borderline commodity product. While, arguably, NotebookLM is the only thing approaching a mainstream consumer product, there's a long tail of free and paid products that do this, and it's competing with about 30 million new podcast episodes a year produced by humans.

Further, the taste agent runs on the one thing no other company can emulate: over a decade of listening history. "The Okee Dokee Brothers are a tell" is a sentence only Spotify can write to you.

In my fifth issue, I introduced a framework for building an AI consumer product in a hostile market. Let's put both sides of Studio to the test.

Q1: What's the counterfactual?

Podcast generator: Fail. From an entertainment standpoint, the alternative is a human podcast. Generally, people listen to podcasts for the person — access to someone otherwise inaccessible (Smartless), as a hang out vibe (Rewatchables), or the unique storytelling the individuals bring to a topic (Planet Money, Acquired). I generated a "making of" film podcast in Studio ("The Cutting Room") and listened to an episode about how Apocalypse Now was made. It felt soulless, like Spotify was strip mining culture and selling it back to me in a package delivered neatly to my phone. Fail.

Taste agent: The alternative is... waiting for my daylist, or Wrapped once a year. Maybe I hassle a friend to make me a playlist. They aren't replacing anything here that they don't do already. It's a complement that builds on their hard won identity of "personalized data is fun, not creepy." Most importantly, Spotify has established trust in this area, that's kind of their thing. The playlist I generated for a road trip produced what my music expert friend called a "very solid mid aughts to mid 10s indie selection." Pass.

Q2: Did they opt in?

Both: Studio is a separate app you have to go out of your way to download. Pass.

Q3: Can your user see the inputs?

Podcast generator: Technically, they show the plumbing, but not the sources. You can watch it retry TTS chunks against a 30-second floor and request shell permissions, but it doesn't tell you what it read or why. For all I know, I listened to 36 minutes of hallucinated bullshit across the three podcasts I generated. Fail.

Taste agent: The user is the input. It's built on your listening history. It's auditable against your memory. It characterized my afternoon account activity—primarily children's music—as "time with kid," not knowing that my account also plays through our Google Home speakers, which my nanny has used during the day for the last month or so. I could tell that the "listening story" it gave me was accurate but recency-biased.
Studio's memory is also stored locally and inspectable. Pass.

Q4: What happens when it's wrong?

Podcast generator: The podcast generator does ask a lot of questions upfront: follow up contextual questions, type of podcast, tone, and voice. And a corrective path exists if it gets something wrong. I switched around topics mid-build and fixed an audio artifact in one of my podcasts by chatting with the agent. Unfortunately, the errors were frequent, sometimes confusing, and often annoying. (The mispronunciation of "Cannes" and "Redux" in the Apocalypse Now podcast was particularly grating.) It's not quite troubleshooting OpenClaw, but it would be jarring for the uninitiated. Soft pass.

Taste agent: Similar to the podcast generator side, correction is built into the conversation. It asks follow-up questions upfront to make sure it has what it needs to get to a helpful output. You can ask it to correct or forget aspects of local memory (although Big Spotify is always listening, so that doesn't change your account data). Pass.

The closing heuristic: would this feature retain users if a compelling alternative existed?

Podcast generator: A compelling alternative exists, and it's free (tens of millions of podcast episodes a year). Two of the three podcasts this produced (both "entertainment" ones) were tough to listen to and actively made me question my Spotify subscription.

Taste agent: There isn't one. Nobody else has the data.

The funny thing is: Spotify is paying top dollar for the alternative to its podcast generator.

Data is sparse, but Spotify revealed their Q1 2025 payouts to podcast publishers topped $100M in a program aimed at pulling video creators away from YouTube. One mid-size horror podcast's quarterly Spotify income went from $45K to $82K after joining.

In the last six months, Spotify also signed a landmark deal with Jay Shetty (up to $100M, in partnership with Netflix) and renewed Joe Rogan's deal (allegedly worth $250M).

Spotify knows what the right person is worth and its talent and exclusivity deals prove it. The podcast generator is the same product, with the key ingredient hollowed out. The taste agent is the opposite — it's entirely about one person, you.

In fairness, Studio is an audio-only Labs experiment next to nine-figure talent deals, not their core strategy. They know their real podcast fight is against YouTube.

While I understand the temptation, Spotify's advantage in AI isn't content generation. A competent vibe coder could wire up Claude Opus with ElevenLabs voice generation API and start generating podcasts in a day, and it would also be worse than a human. Their advantage is knowing that I play Okee Dokee Brothers instead of Baby Shark on repeat (what can I say, I still have nostalgia for 00s/10s stomp clap).

Spotify knows this. Their landing page and FAQ even say so. The product just doesn't act like it yet with its heavy-handed push toward podcast generation.

Even the build gives it away: the assistant pattern is OpenClaw's, the harness is Pi, the model is rented and unnamed. Maybe they did some prompt engineering to guide the podcast features, but the only thing in Studio that Spotify owns is your listening history — the part that's hidden away.

I don't love Studio in its current form (a mobile app would be preferable), but I'd love to see the capabilities integrated into its parent app.

Send me questions, feedback, and cocktail recipes:
[email protected].

_{Icons made by Icongeek26 from www.flaticon.com.}