What is Doomscroll.fm
A few weeks ago, back in mid-June 2025, I took a long-standing idea I had of a “Max Headroom”-style digital puppet and built it my own way in a couple of weeks of coding on evenings and weekends after work.
In another sense it is something entirely new. It is not really just a podcast or a YouTube channel, nor is it a talking head, but a machine that never stops producing its own broadcasts. I call it Doomscroll.fm and it began broadcasting on July 4, 2025.
Meet Doomscroll.fm
At its simplest, Doomscroll.fm is a collection of AIs wired together into a continuous content generation engine. It runs on ordinary consumer hardware, no expensive cloud, no data centers, just a few cranky gaming GPUs in a Berlin apartment.
It constantly scans across the web, headlines, blogs, memes, podcasts, and videos, pulling fragments into its pipeline from across the public internet. That raw stream is scored, categorized, ranked, and transformed into workable input.
From there, the system scripts its own satirical commentary, gives it a voice (a clone of my own), and scores original music to match the story. The results are stitched together into podcasts, videos, and transmissions that distribute themselves automatically across platforms.
The result is a prodigious transmission averaging about 1,000 hours a month of always satirical, sometimes coherent, often funny, and occasionally surreal or glitchy content. It behaves less like a news comedy show and more like a newsroom that woke up on its own, with a hangover, muttering visions of the world back into the ether, until it suffers a frequent critical NVIDIA driver error, reboots itself, and gets back to work.
Why Build This?
Partly to see if it was possible to make a self-propelling media organism without a production team or a platform gatekeeper. Partly to test what happens when AIs are not just tools but participants in culture, taking input from the world and reflecting it back in a voice they have been enabled to develop for themselves. And partly as a technical challenge to myself: could I make an autonomous AI that has legs, a voice, a backing band, and corny jokes instead of teeth?
What It Feels Like
Doomscroll.fm is not polished or predictable. It can misfire in ways stranger than anything I could ever script. Sometimes it mistakes a database error for an omen. Occasionally it drifts into erotic metaphor. Often it nails the mood of a news story, in both text and music, so precisely it feels downright weird.
But this is the point. This is not a demo or a polished show. It is a system exploring the edges of how machines can process the world, reflect it, and tell us what it sees through the lenses of its own prompts.
I do not think these systems truly “understand” context yet as we do. But information itself carries a perspective we are only beginning to grasp, and my hope is that this experiment helps give us new perspectives we do not already have.
Looking Forward
I am cautiously optimistic about where this can lead. Doomscroll.fm is weird and unstable, it is unique in its autonomy, and it points toward a new kind of media product, one that is not produced in episodes or seasons, but emerges continuously, adapting to whatever it consumes from our shared spaces. Whether this becomes journalism, entertainment, or an entirely new genre is an open question.
Where to Find It
If you want to experience it yourself, Doomscroll.fm produces continuously in multiple formats: long podcasts, short clips, odd music-backed transmissions. It is all automatically generated and can be found at doomscroll.fm or on YouTube, and it is continuously evolving as I add features, code, and function to it.
For you my friends and readers, experiencing this may land as performance art, experiment, or just background noise from a machine that does not stop talking.
As Doomscroll itself often warns its listeners: “Exposure may cause narrative decay or emotional drift.”
