Back to articles
Share:
AI & Technology

How the AI Music Is Made

June 3, 2026
5 min read

Anthony R Quigley

Writer, builder, and explorer of ideas at the intersection of technology, systems, and human experience.

Quigley Dynamics is a self-running radio experiment. The songs, station IDs, fake advertisements, and weekly podcasts are generated by software, then assembled into a listening surface that behaves more like a station than a static playlist.

Two Streams With Different Rules

The system currently runs two channels. MAINFRAME leans nocturnal: dub techno, trip-hop, dark orchestral textures, and slow cinematic pressure. PARITY PARADE is brighter and stranger: roaring-twenties jazz, electro-swing, brass stabs, upright bass, and retro-futurist club energy. They share the same machinery, but each station has its own style grammar.

Prompts Become Tracks

Each track begins as a description of a vibe rather than a reference to a real artist. The generator is asked for instrumentation, tempo, mood, rhythmic feel, and mix character. Those prompts are varied across a style spectrum so the library can grow without collapsing into one sound.

The raw audio is only the first pass. After generation, each file is normalized for consistent loudness, trimmed into a usable radio asset, and prepared so it can move cleanly through the live player, the song board, and the podcast renderer.

The Radio Layer

A radio station needs more than songs. The system also creates short DJ moments, station identifiers, and fictional ads. Those elements are treated as first-class audio assets, then scheduled between tracks so the stream has texture: a voice in the booth, a strange sponsor, a signal marker, then another song.

The Vote Signal

The public song board is the feedback loop. Listeners can play tracks, upvote what should stay in rotation, and downvote what should fade out. The board is live at anthonyrquigley.com/vote. Those votes feed back into selection: popular songs get more chances, and disliked songs lose weight.

That vote signal also matters for the podcast. When the weekly show is rendered, the pipeline favors new material and folds in catalog tracks that listeners have pushed upward. The result is not just a chronological dump of generated files; it is a programmed episode shaped by the audience.

The Weekly Podcast Pipeline

Once the station has enough fresh material, the renderer stitches a continuous show: songs, DJ breaks, station IDs, and ads. Each station gets its own episode and feed art. The podcast is the archive version of the live system, a weekly snapshot of what the machine made and what listeners wanted to hear again.

Why Build It This Way

The point is not to pretend the machine is a human band or a human host. The interesting part is the loop: generation, curation, playback, voting, publishing, and then generation again. It is a small test of what happens when a creative system is allowed to run continuously and the audience becomes part of the control surface.