Vestigim

Over the past five months, two of the three major labels settled with the AI music companies they sued in 2024 and announced licensed platforms with them. Sony is still in court. In January, music publishers filed a copyright case against an AI lab asking for more than three billion dollars. Spotify, meanwhile, spent the fall building AI disclosure into song credits and removing tens of millions of spam tracks.

That’s the industry conversation about AI and music: generation, infringement, licensing, flood. Prompt in, song out, lawsuit after. And I understand why it dominates — livelihoods are at stake. But as someone who spent years of my life in studios and on stages before I worked in technology, I think the whole fight is over the wrong asset.

A music catalog has priced exactly two things for a century. Masters — the recordings. Publishing — the compositions underneath them. Both are records of what an artist already made: finished, and fully valued. What no one has ever owned is the layer above the output — the judgment that chose it — and the layer beneath it — the craft that produced its sound. Vestigim is my work on those two layers.

Two more layers of the catalog

Vestigim maps an artist’s judgment, sound, career, and world into distinct, ownable, licensable layers that didn’t exist as assets before. The judgment, career, and world resolve into one living picture — the Vestigim Graph. The sound — how an artist actually gets their tone — becomes Machine Sounds. Together they’re a third and fourth layer of the catalog: not what the artist made, but how they decided and how they sounded, as things a machine can read.

This is a forward bet. Catalogs still price only masters and publishing today. But in 2026, machines are becoming the main consumers of an artist’s sound and an artist’s judgment — and that’s what turns a disposable preset, or a pattern of choices, into an asset worth owning. When machines are the audience, how you decide and how you sound stop being byproducts of the work. They become the work.

A memory of judgment

The Vestigim Graph is the closest thing to a living memory of an artist: their press, their charts, their scene, their collaborators, their career and life — all of it organized around one thing, the way that artist makes decisions. And that’s the part that matters, because it isn’t a model of what an artist tends to produce. It’s a model of what they chose — and, just as importantly, what they turned down, and why.

That second half is where the real signal lives. What an artist throws away says more than what they keep — the take they cut, the mix they held, the feature they walked away from. Vestigim reads the work itself, reads the choices behind it, and only asks the artist directly when the evidence runs out. Everything it believes comes with a source. It doesn’t assert; it shows its work. That’s the difference between a memory a system can trust and a guess it can’t.

Not who they were — how they’d decide next

Because that memory holds the reasons and not just the results, it can be run forward. Point it at where the scene and the moment are heading, and it can surface the moves an artist would lean into, reject, or want to explore next — not a prediction of the future, but their own taste, applied to a field that hasn’t happened yet.

A queryable model of an artist’s judgment is a genuinely new object, and the uses stack up fast. A real-time read on how an artist’s instincts move against the market. A licensed check other AI systems can call to ask whether an output actually sits inside an artist’s taste, or just borrows their name. A signal an investor can weigh when valuing a catalog. A way to reconstruct a historic artist’s decision-making from the evidence they left, for scholarship or a sanctioned live experience. Each of these is something the industry has wanted and never had a way to own.

The craft, captured — not the preset

The other layer, Machine Sounds, is an artist’s authenticated sound-craft — their tones, their moves, the way they engineer a sound, made ownable and reproducible. Not a frozen preset, but the living behavior behind it.

The reason this is hard is worth stating plainly. A real piece of studio hardware doesn’t have a “setting” you can copy — it’s a physical thing where every knob changes how every other knob behaves, across everything you feed into it. Most tools capture a snapshot of one setting and call it done. Machine Sounds captures the whole instrument: its actual behavior across its entire range, measured from the real unit, so what you license sounds like the hardware because it is the hardware, modeled faithfully — with a clear line of provenance back to the physical device. It captures a sound; it never invents one. That’s exactly what makes it something you can license rather than fake.

One graph, and a firewall

The two layers are joined, but only in one direction, and that direction is the whole ethic of the thing. The Graph learns from Machine Sounds — an artist’s sonic choices are evidence of their judgment — but Machine Sounds never learns from the Graph. Sound informs the picture of the artist; it never gets mistaken for the artist. Only genuine decisions carry an artist’s taste. Everything else — the sound, the market, the press, the whole surrounding world — is context that judgment reacts against, never a stand-in for it. That rule is built into the structure, not bolted on as a promise. It’s what lets the system take in an entire industry’s worth of noise around an artist without ever confusing the noise for the person.

On the rights-holder side of the line

There’s a reason the architecture is this careful, and it’s the same reason the lawsuits exist: in the age of AI, nobody can prove what happened to a piece of music — whether a work trained a model, whether a track was made by a person or a system, whether a voice was licensed or lifted. Unauthorized use became common precisely because there was never an audit trail.

Vestigim is built on the other side of that line. It does not sell AI-generated music. Every asset is human-authored, with provenance back to a real work, a real decision, or a real instrument — the licensed inputs that any downstream system has to license to operate honestly. The models are living because they keep updating, not because they resurrect anyone. We map a practice — judgment, craft, and how they evolve — not a person. We don’t clone the artist. We license their judgment.

I don’t think AI in music is going away, and I don’t think fighting it as a category works — the settlements of the past few months suggest the industry has quietly reached the same conclusion. Even the sharpest artist statements against AI exploitation have been careful to say the technology, used responsibly, has enormous potential. The real dividing line was never AI versus no AI. It’s whether the tools answer to the people who make the music. Masters and publishing gave artists ownership of what they made. The next two layers give them ownership of how they made it — and that’s the standard I’d want as a musician, so it’s the one I’m building as a technologist.