Neo-Tokyo 2045: what happens when you ask an AI for a city that doesn't exist yet

Spotify knows what you already like. That's the whole product. Feed it five Tame Impala tracks and it hands you a bucket of psych-rock adjacent things — competent, safe, recognizable. The algorithm lives inside your taste like a polite house guest who never opens the cupboards.

That's useful. It's also why most of us hit a wall around our second month on the platform. You want something specific. Not "more of what I like." You want "songs that sound like walking through Shinjuku in 2045, after the rain, when the neon is scattered across the puddles and nobody is really looking at anyone."

Spotify can't do that. The algorithm has no hook for a scene that doesn't exist. It can match vibes it has already seen, which means anything future or unbuilt or hyper-specific falls through the sieve.

So we built Playgen.

The prompt

"Make a playlist to feel like walking in Tokyo in 2045."

That was it. One sentence. Playgen turned it into a twenty-track Spotify playlist in about twelve seconds. You can listen to it here:

→ Neo-Tokyo on Spotify

It opens with Yellow Magic Orchestra's TECHNOPOLIS (the 2018 Bob Ludwig remaster — because details matter), ambles through city-pop, slides into late-night synthwave, and ends somewhere you didn't expect. If you closed your eyes during the first track and imagined wet pavement and a chrome overhang, the playlist did its job.

How it actually works

There's no magic. There are three steps.

1. The LLM picks the tracks

Your prompt goes to a large language model. It isn't searching a database. It's reasoning about what a scene sounds like — what era, what instrumentation, which artists, which specific tracks. It returns a structured list: twenty objects, each with an artist and a title.

This is the part that Spotify's recommender literally cannot do. Spotify recommends based on user-taste graphs: people who liked X also liked Y. A language model reasons from the semantics of the prompt — it knows Tokyo, it knows 2045, it knows city-pop, it knows what future-cities in fiction sound like. It can bridge concepts.

2. Find them on Spotify

Having an artist and a title isn't enough — you need Spotify's canonical track URI. So Playgen fires off twenty parallel search queries against the Spotify Web API, taking the first hit for each. Over-requests a few extra so misses don't leave holes. If the primary LLM returns something weird, we retry once with a fallback model.

3. Publish as a real playlist

The matched URIs go into a public Spotify playlist on a shared "Playgen" bot account. You get a clickable Spotify link that works on web, desktop, iPhone, Android, Sonos, anything. Save it to your library, share it, pretend you made it. We don't care.

Why this is different

The interesting thing isn't the AI. The interesting thing is what you can now ask for.

A short tour of prompts people on Playgen have actually run:

Songs a Berlin bartender would play at 3am.
Soundtrack for writing a breakup text you'll never send.
Japanese city pop meets French house.
If Tame Impala produced a Bollywood soundtrack.
Bangers in 7/8 time.
What a mass at Notre-Dame would sound like if it was techno.
Brazilian funk meets Scandinavian jazz.
If Chopin was an ambient producer in 2025.

Every one of those is a scene, not a taste. Try typing any of them into Spotify's search bar. You'll get nothing — or worse, you'll get what the algorithm thinks is "close enough," which is a watered-down version of an entirely different idea.

Playgen is the tool you reach for when the prompt in your head is more cinematic than categorical.

Try it

It's free to start — you get one trial credit on sign-up, and after that it's about $0.10 per playlist depending on the pack. No subscription. Credits don't expire.

→ Try Playgen now

If you make something weird and specific, we'd love to see it. Tag @gabinfay on X or just drop the Spotify link somewhere public. That's how we find the prompts worth showcasing.

Now go build a city that doesn't exist yet.