AI Made Me a DJ Set. Here’s What It Taught Me About Digging

We've all become aware of Artificial Intelligence making music, but have we considered what It will do to the overall ecosystem of record digging, listening, music discovery, and sequencing DJ sets?

Take for example, Spotify. Spotify has been using machine learning to predict our taste for years. Generative AI is now giving us a way to actively converse with those systems about what we want. Most people are drawn toward this music discovery tool for background music to their lives. It's easy to turn on, and It helps them find tracks they enjoy listening to.

Spotify is now explicitly moving further into generative AI. In 2026, Spotify described its strategy as moving from “recommendation” to “generation”, saying it is building a proprietary “Large Taste Model” based on billions of behavioral signals. (source)

AI Is Very Good at Extrapolating Taste

So I gave ChatGPT a screenshot of one of my SoundCloud playlists and asked it to recommend more tracks. Then I asked it to arrange them into a DJ set: something that could build, develop, reach a peak and eventually hand over to the next DJ.

I wanted to compare AI music discovery with the old fashioned practice of record digging. It gave me more or less similar producers to the tracks that I already listened to, but they lacked any type of storytelling or curveballs. It was a 'safe' recommendation, tracks that fit well together, but lacked creativity or risk taking.

I realized I wasn't necessarily looking for more dubby/psychedelic tracks. I was looking for the feelings of these records I'd chosen: the groove, slightly strange percussions, the label's history it came from, the perfect layering right moment kind of edge.

And that distinction made me rethink what we actually mean when we talk about digging for music.

If AI can reproduce the informational part of digging, what is the very thing we value when going to a record shop instead?

Can AI replace music discovery?

AI is extremely good at extrapolating taste. It can look at ten records and map the neighborhood around them, but there’s a potential problem. It tends to assume that the thing you’re looking for is inside the neighborhood you’ve already described. A great record shop digger doesn’t necessarily do that.

I'd always assumed there was something more valuable about digging for music yourself. Spending hours in record shops, going down a label rabbit hole, buying a record because the sleeve caught your eye, without knowing whether you'd actually like it.

Streaming platforms have been doing a version of this for years: looking at what you've listened to and finding other music that statistically resembles your behavior. AI takes this somewhere else. You can describe why you like something, talk about texture, atmosphere, rhythm, context, a particular era or scene. You can ask it to interpret the relationship between records rather than simply identify similarities.

	Spotify / Streaming Platforms	AI	Bandcamp/SoundCloud	Record shop
Discovery	Recommendation	Interpretation	Exploration	Encounter
Starting point	Your history	Your prompt	Artists/labels/scenes	Physical environment
Main strength	Scale	Context	Depth	Serendipity
Risk	Reinforcement	Safety	Rabbit holes	Getting lost

(scroll left to view whole chart)

It can take: "I like this" and turn it into: "Here are 50 things you might also like." It tends to assume that what you're looking for is somewhere inside the boundaries of the taste you've already described. Whereas, a great record shop digger doesn't necessarily work that way. They might hand you something that doesn't fit at all. And there's so much more Information handed down to you when you grab the physical copy.

Why DJs Still Dig for Records

You walk into a shop looking for deeper late night minimal tracks and someone behind the counter puts on a record from 1997 that technically makes no sense. You don't like the first 90 seconds or it feels off from what you came in for. Then you hear a hi-hat, or a bassline, or some strange production technique, and you understand your original taste differently and the connection between the two. The discovery isn't: A to B. It's now A to something unexpected to a change in your understanding of A to B, C, D...

That's much harder to reproduce, and something AI isn't doing for us now. A record shop isn't a recommendation engine like streaming platforms or Chat GPT. The information in a record shop is much richer. It's densely providing information in the sleeve, the typography, the label, the pressing, the distributor, the physical bins, the handwritten notes, the records sitting next to it. It's the person behind the counter, the person who bought it before you. It's a community space and the lineage that came before you.

You might pull out a record because you like the artwork, notice it's on a label you've never heard of, look through the rest of the catalogue, discover three artists, find out that one of them released something ten years earlier under another alias. This process is the art of the dig, and holding a piece of history - the lineage. This is what makes digging different from receiving recommendations from machines.

Good taste isn't only information acquisition, it's relationship formation.

Historically, finding music required work. You needed knowledge, money, time, access, friends, record shops, magazines, radio, DJs, labels and travel. You had to physically go somewhere, to ask someone, to listen to things you weren't necessarily looking for. Maybe you received a flyer from someone that lead you to discover something totally weird.

The Wasted Time Is Not Wasted

You can spend two hours in a record shop and buy nothing, from the perspective of an optimization engine, that's failure, from the perspective of digging, it might be the entire point.

You're looking, listening, picking something up, putting it down, talking to someone, getting distracted. following a label you don't know, smoking a cigarette, finding a record that makes absolutely no sense, lost and then found again in conversations. The wasted time is not wasted.

"If you play a sound, you have to pick out … the most important note that fertilizes the sound.” - Miles Davis

The paradox of AI and taste

If everyone has access to an AI capable of producing a competent 50 track playlist, the playlist itself becomes less valuable. The real value then becomes: why these 50?

A DJ, record shop owner, collector or obsessive listener of course possess information, and they also possess a history of choices. Their taste has been shaped by thousands of encounters. Records they disliked, misunderstood, heard at 5am, someone insisted they listen to, bought for the artwork and eventually fell in love with, inherited it from someone they admire, recommended by a DJ they heard play it. That history gives their recommendation historical weight.

Personally when I travel to different countries, I want to pull records that hold the sounds of the region. In Brazil, I grabbed a traditional folk record from the 60-70's, something more avant garde and punk, then of course a Bossa Nova record.

This eclectic selection enhanced my memories of my trip, and I'll remember certain details because of the sounds that I otherwise wouldn't have.

Perhaps that's why the human element of digging becomes more interesting. The more easily we can access everything, the more valuable a particular point of view becomes.

What Spotify and AI Can't Give You

The purpose of digging isn't simply to find records. It's to develop the ability to find records. After years of digging, you start recognizing labels, catalogue numbers, distributors, mastering engineers, sleeve designers, aliases and eras. You see a record you've never heard and somehow have a feeling about it.

What surprised me most wasn't that AI knew the artists - I expected that. What it couldn't do was make me want something I hadn't yet learned how to ask for.

Maybe that's the dividing line. Not human versus machine, vinyl versus streaming, but retrieval versus discovery.

I don't need a record shop to tell me there are 500 records similar to the one I'm holding. I can ask a machine. What I still want from the shop is the possibility that someone puts a record in my hand and says: I don't know if this is what you're looking for, but listen to this.

The important part isn't that they know what I like, it's that they might change what I like, we can talk about it, and the music culture lives on.