#1952: Why We Built a 24/7 AI Radio Station

We turned our 1800-episode archive into a continuous AI-powered radio stream. Here’s the tech stack and the philosophy behind it.

0:000:00
Episode Details
Episode ID
MWP-2108
Published
Duration
20:42
Audio
Direct link
Pipeline
V5
TTS Engine
chatterbox-regular
Script Writing Agent
Gemini 3 Flash

AI-Generated Content: This podcast is created using AI personas. Please verify any important information independently.

The Problem with a Deep Archive

When you have over 1,800 episodes, you have a massive wall of content. For a new listener, it’s daunting—like being handed the keys to the Library of Congress with no map. Even for the creators, navigating the nuances of a deep dive from two years ago can be a challenge. This raises a fundamental question: in an age of on-demand streaming and infinite scroll, how do you make a vast archive accessible and engaging without overwhelming the listener?

The Lean-Back Experience

The solution we explored is a return to a classic format with a modern twist: a 24/7 internet radio station. This isn't about replacing the podcast feed; it's about complementing it with a "lean-back" experience. The core idea is rooted in the psychology of choice. Constant decision-making leads to fatigue. Sometimes, you don't want to browse a catalog and commit to a specific topic for 40 minutes. You just want to tune in and let the stream take the lead. This creates an ambient presence, similar to having the television on in the background while you go about your day. It removes the mental cost of commitment and opens the door for serendipitous discovery.

The Technical Stack: Icecast and Liquidsoap

Building this station is surprisingly elegant and accessible. The entire operation runs on a virtual private server, like a DigitalOcean droplet, and relies on two open-source powerhouses:

  • Icecast: This is the server component, acting as the digital transmitter. It handles the actual distribution of the audio stream to listeners. When you connect to the stream, your player pulls the audio from an Icecast mount point over HTTP.
  • Liquidsoap: This is the programmable engine, the "DJ" of the station. Unlike a simple media player, Liquidsoap uses a scripting language to define the broadcast logic. It shuffles the entire archive, handles transitions between episodes, and normalizes audio levels to prevent sudden volume changes. The script can also manage different audio sources, such as a folder of station jingles, and insert them at programmed intervals. It even includes "fallback" logic to ensure the stream never goes silent if a source fails.

Metadata and the "Water Cooler" Effect

A key feature of modern radio is metadata—the song title that appears on your car's display. This station achieves the same effect. Liquidsoap reads the ID3 tags from the audio files (title, episode number, description) and sends updates to Icecast, which broadcasts them alongside the audio. This means a listener can see exactly which episode is playing in real-time.

This creates a unique, synchronized experience. Unlike a podcast where every listener is on a different timeline, a live stream means everyone hears the same word at the same moment. It fosters a sense of a virtual community, a "water cooler" effect where listeners are sharing a moment, even if they are miles apart.

Resurrecting the Long Tail

For creators, this model is a powerful tool for unlocking the value of their "long tail" content. In a traditional podcast feed, old episodes quickly sink to the bottom, rarely heard by new listeners. A continuous stream gives every minute of content an equal chance of being discovered. A listener might tune in and hear an episode from seven hundred shows ago about automated prompt engineering—a topic they would never have sought out but are now enjoying. This "forced discovery" feels less like a chore and more like a gift, keeping the archive alive and relevant.

The Future: The Infinite Broadcast

The logical next step is an "infinite broadcast." The current system streams a pre-recorded archive, but the framework is already in place for something more dynamic. Imagine a Liquidsoap script that, instead of pulling a static file, calls an API to generate a fresh segment of audio in real-time. This could include location-specific weather reports, news breaks from minutes ago, or entirely new AI-hosted content woven into a seamless, never-ending stream.

Practical Hurdles and Ownership

For other creators interested in this model, the steepest learning curve is the Liquidsoap scripting language, which is functional and requires understanding how sources and transitions work. However, a strong community and templates are available to help. A significant advantage for this podcast is that it owns all its content. Unlike traditional radio, which faces complex music licensing and royalty fees (e.g., ASCAP, BMI), this station streams only original talk and AI-generated audio. This vertical integration—producing, generating, and broadcasting in-house—cuts out every middleman except the server host, making it an efficient and self-contained media ecosystem.

Downloads

Episode Audio

Download the full episode as an MP3 file

Download MP3
Transcript (TXT)

Plain text transcript file

Transcript (PDF)

Formatted PDF with styling

#1952: Why We Built a 24/7 AI Radio Station

Corn
You know, Herman, I was looking at the website this morning and it hit me just how much digital real estate we are occupying now. We have over eighteen hundred episodes just sitting there in the archives. It is a massive wall of content. If a new listener shows up today, where do they even start? It is like handing someone the keys to the Library of Congress and saying, have fun, hope you find something cool.
Herman
It is a daunting task for any listener, honestly. Even for us, sometimes I forget the specific nuances of a deep dive we did two years ago. But that is exactly why today’s prompt from Daniel is so timely. He is asking us to pull back the curtain on the new My Weird Prompts twenty-four-seven internet radio stream. We have officially moved beyond being just a podcast feed where you pick and choose. We are now a continuous broadcast.
Corn
It feels a bit retro, doesn't it? Like we are circling back to the golden age of radio, but with a massive AI-generated twist. And by the way, for everyone listening to this specific breakdown, today’s episode is actually powered by Google Gemini three Flash. It is fitting that the model writing the script is helping us explain the machine that keeps the stream alive. I want to start with the obvious question, though. Why? In an age where everything is on-demand, why would we build something that takes away the listener's ability to hit play on exactly what they want?
Herman
Herman Poppleberry here, and I have been waiting to nerd out on this. The "why" is actually about the psychology of choice. We live in a world of infinite scroll and decision fatigue. Sometimes, you don't want to browse a catalog of two thousand episodes. You just want the vibe. You want to tune in and let the stream take the lead. It creates this ambient presence. It is the difference between going to a cinema to watch a specific movie and just having the television on in the background while you go about your day.
Corn
I get that. There is a comfort in the "lean-back" experience. But let’s get into the guts of this thing. When we say "internet radio station," I think some people imagine a massive studio with blinking lights and a giant antenna on the roof. In reality, this is all living on a server, right? What is the actual stack making this happen?
Herman
It is surprisingly elegant and, honestly, quite accessible if you are comfortable with a bit of scripting. The core of the operation is a combination of two open-source powerhouses: Icecast and Liquidsoap. We have this running on a DigitalOcean droplet, which is basically just a virtual private server. Icecast is the server component. You can think of it as the digital transmitter. It handles the actual distribution of the audio stream to the listeners. When you go to radio dot myweirdprompts dot com, your browser or media player connects to the Icecast mount point, and Icecast pumps the bits to you over HTTP.
Corn
Okay, so Icecast is the pipes. But who is the plumber? Who is actually deciding that Episode four hundred twelve should play after Episode nine hundred?
Herman
That is where Liquidsoap comes in. Liquidsoap is the engine, or the "DJ," if you want to stay with the radio metaphor. It is an incredibly powerful audio toolkit. Unlike a simple music player, Liquidsoap is programmable. You write scripts for it. Our script tells Liquidsoap to look at our entire archive of eighteen hundred plus episodes, shuffle them, and then prepare them for streaming. But it does more than just shuffle. It handles the transitions. It ensures that the audio levels are normalized so you don't have one quiet episode followed by me shouting about battery chemistry at twice the volume.
Corn
I think the listeners would appreciate the normalization, Herman. Your shouting is an acquired taste. But wait, if it is just a script, how does it handle things like station IDs or the little jingles we have? Does it just treat them as files in the folder?
Herman
Well, not exactly—I mean, it treats them as specific sources. In the Liquidsoap script, you can define different sources of audio. You have your "archive" source, which is the big folder of episodes. Then you have your "jingle" source. You can program logic that says, "Every three episodes, play one random file from the jingle folder." You can even set up "fallback" logic. If for some reason the main playlist fails, Liquidsoap can automatically switch to a single emergency backup file so the stream never goes silent. It is incredibly robust.
Corn
This is what fascinates me. We are essentially mimicking the workflow of a professional radio station, but with zero human intervention once the script is running. It is a "set it and forget it" broadcast. But let's talk about the stream itself. This isn't like a podcast where you download a file and your phone knows it is forty minutes long. If I tune in mid-sentence, I’m just... there.
Herman
That is the magic of the bitstream. It is a continuous flow of data. When you connect, you are catching the stream at whatever point it is currently at. There is no "start from the beginning" button on a live stream. This creates a sense of synchronized arrival. If you and I both tune in at ten o'clock in the morning, we are hearing the exact same word at the exact same time. In a world where digital experiences are increasingly isolated and personalized, there is something almost communal about knowing other people are listening to the same deep dive right now.
Corn
It’s the "water cooler" effect, even if the water cooler is virtual. But I have to ask about the metadata. One thing I love about modern radio is seeing the song title on my car dashboard. How do we get the episode titles to show up when someone is listening to the stream?
Herman
That was one of the fun technical hurdles. Icecast supports metadata updates, and Liquidsoap can extract that information from the audio files themselves. As Liquidsoap processes an MP3, it reads the ID3 tags—the title, the episode number, the descriptions—and it sends a metadata update to Icecast. Icecast then broadcasts that metadata along with the audio. So, if you are using a player like VLC or even just a modern web browser, you’ll see "Episode fifteen hundred: The Future of Carbon Capture" pop up right as the audio starts. It makes the archive feel alive and searchable, even though you aren't the one doing the searching.
Corn
I’m curious about the resource usage. Running a twenty-four-seven stream sounds like it would be heavy on the CPU or the bandwidth. Are we melting that DigitalOcean droplet?
Herman
Not at all. That is the beauty of Icecast and Liquidsoap being optimized for this. Since we aren't doing heavy live re-encoding of the audio—we are mostly just streaming pre-encoded MP3s—the CPU usage is actually very low. The main cost is bandwidth, but even that is manageable. Audio bits are relatively small compared to video. We can serve hundreds of concurrent listeners on a modest server without breaking a sweat. It is much more efficient than people realize.
Corn
So, we have the technical stack down. Icecast is the transmitter, Liquidsoap is the DJ, and it’s all running on a virtual server for the price of a few cups of coffee a month. But let’s shift gears to the listener experience. I’ve been spending some time just leaving the stream on while I’m working, and it’s a totally different feeling than "podcasting." Why does it feel so different?
Herman
I think it is the lack of commitment. When you pick a podcast episode, you are making a conscious choice to spend forty minutes with a specific topic. There is a mental "cost" to that. If the topic doesn't grab you in the first five minutes, you feel like you made a bad choice. With the radio stream, that pressure is gone. You just turn it on. If you join in the middle of a conversation about geopolitical strategy in the Middle East, you just sort of float with it. You aren't committed to the whole arc; you are just enjoying the current moment of the "vibe."
Corn
It’s like a discovery engine. I found myself listening to an episode from back in the seven hundreds the other day—something about the early days of automated prompt engineering—and I realized I had completely forgotten some of the points we made back then. I would never have gone looking for that episode. It was "forced" discovery, but in a way that felt like a gift rather than a chore.
Herman
That is a huge point for creators. In the traditional podcast model, your "long tail" content—your old episodes—usually just goes to die. They sink to the bottom of the feed and nobody ever sees them again unless they are a completionist. A twenty-four-seven radio stream resurrects your entire history. It gives every single minute of content you’ve ever produced an equal chance to be heard by someone new. For a show like ours with nearly two thousand episodes, that is a massive unlock for our intellectual property.
Corn
It also changes how we think about the "station" identity. We aren't just a show anymore; we are a destination. You don't just "listen to My Weird Prompts," you "tune into My Weird Prompts." It feels more permanent. But what about the future of this? We are talking about AI-generated content. If we can generate scripts and audio indefinitely, does the concept of an "episode" even matter anymore? Could we just have a "living" stream that never repeats and never ends?
Herman
That is the frontier. Right now, we are streaming our archive, which is a collection of discrete episodes. But the logical next step—and something we are already seeing hints of in the industry in this year of twenty twenty-six—is the "infinite broadcast." Imagine a Liquidsoap script that, instead of just pulling an old file, calls an API to generate a fresh segment of audio in real-time. You could have a weather report that is actually accurate to the listener’s location, or a news break that happened five minutes ago, all woven into a continuous stream of AI-hosted content.
Corn
That sounds both incredible and slightly terrifying. It’s like the Truman Show, but for podcasts. But let’s look at the practical side for anyone else who might want to do this. You mentioned that this is simpler than it sounds. If a smaller podcaster wanted to set this up, what are the actual hurdles? Is it the scripting?
Herman
The scripting is the steepest part of the learning curve. Liquidsoap uses its own language, called "lang," which is a functional programming language. It isn't like writing a simple bash script. You have to understand how sources and transitions work. But there is a great community behind it, and plenty of templates. The other hurdle is just the server management—making sure your Icecast server is secure and that your mount points are properly configured. But honestly, compared to the complexity of building a custom app or managing a high-traffic website, an internet radio station is a very contained project.
Corn
And no licenses? I know traditional radio is a nightmare of FCC regulations and broadcast licenses. Does the internet version bypass all that?
Herman
Well, you still have to worry about music licensing if you are playing copyrighted songs. That is the big one. If you play Taylor Swift on your internet radio station, you need to be paying royalties to organizations like ASCAP or BMI. But for us, since we own all our content—it is our voices and our scripts—we don't have those hurdles. We are our own record label and our own broadcaster. That is why this model is so perfect for talk-heavy podcasts or AI-generated shows. We own the rights to every single bit we stream.
Corn
It’s the ultimate vertically integrated media company. We produce the ideas, the AI generates the scripts, the TTS generates the audio, and now our own server broadcasts it twenty-four-seven. We’ve cut out every middleman except the server host.
Herman
And even the server host is just providing the raw iron. We are the ones who built the "station" on top of it. What I find really compelling is how this affects the relationship with the listener. We’ve talked about the "ambient company" aspect. Think about people who work lonely jobs—truck drivers, night shift workers, developers staring at code for ten hours a day. Having a stream that is "always on" provides a sense of continuity. It is a presence that exists in the world whether you are listening to it or not. There is something deeply human about that, irony of the AI-generation aside.
Corn
It reminds me of the old "numbers stations" or those lo-fi hip-hop streams on YouTube. They become a landmark in the digital landscape. You know they are there. You can count on them. I think for My Weird Prompts, moving into this "always-on" phase is a reflection of how much the world has changed by April twenty twenty-six. We don't just consume media anymore; we inhabit it.
Herman
That is a great way to put it. We are creating an environment. And technically, this enables some really interesting experiments. For example, we could do "live" takeovers. Liquidsoap allows you to "punch in" with a live source. I could be sitting here in my den, hit a button, and suddenly I’m broadcasting live to the radio stream, overriding the archive. When I’m done, it smoothly fades back into the shuffle. It allows for a hybrid of scheduled and spontaneous content that a traditional podcast feed simply cannot do.
Corn
So if you have a sudden, urgent thought about the latest open-source LLM release, you don't have to wait for the next episode drop. You can just... go live.
Herman
Well—again, I shouldn't say exactly—but that is the capability. It turns the podcast into a platform. It makes the distribution as dynamic as the content creation. And when you think about the direction AI is moving, dynamic is the keyword. Everything is becoming real-time. The delay between a prompt being sent and a response being generated is shrinking. The delay between an event happening and an AI discussing it is shrinking. Radio is the natural home for that kind of speed.
Corn
I’m also thinking about the data we get from this. With a podcast, you get "downloads." You know someone grabbed the file, but you don't really know if they listened to the whole thing or where they dropped off, unless you have specific platform-side analytics. With an Icecast stream, you can see exactly how many people are connected at any given second. You can see the "tuning in" and "tuning out" in real-time.
Herman
It gives you a much more granular look at engagement. You can see which episodes or topics cause a spike in listeners or which ones make people disconnect. It is a much tighter feedback loop. Plus, it allows for a different kind of monetization. Instead of just baked-in ads, you could have "live" ads that are injected into the stream at specific times, regardless of which episode is playing. It opens up a lot of possibilities for how we sustain this whole operation.
Corn
Speaking of sustaining, we should probably mention the practical takeaways for our listeners who are curious about this. If you want to experience this yourself, go to radio dot myweirdprompts dot com. It is live right now. It is probably playing an episode about robot ethics or the history of the steam engine.
Herman
Or maybe that one where I tried to explain the entire history of the semiconductor industry in fifteen minutes. That was a marathon. But the takeaway for creators is this: don't let your archive sit in a graveyard. If you have a body of work, a twenty-four-seven stream is an incredibly low-cost, high-impact way to keep it relevant. It turns your past work into a continuous service.
Corn
And it doesn't require a massive team. It’s a weekend project for a curious nerd with a DigitalOcean account. I think that is the most inspiring part. The tools of mass communication have been truly democratized. We are running a global radio station from a server that costs less than a lunch special.
Herman
It’s the "simplicity paradox" Daniel mentioned in his notes. On the surface, it’s just audio over HTTP. But when you combine it with the logic of Liquidsoap and the scale of our AI-generated archive, it becomes something much more than the sum of its parts. It becomes a living, breathing entity. I’m really proud of what we’ve built here. It feels like a milestone for the show.
Corn
It really does. Reaching eighteen hundred eighty-six episodes is one thing, but making them all accessible and "alive" in a continuous broadcast is another. It’s a testament to the power of these tools. And honestly, it’s just fun. There is a certain joy in hitting a URL and hearing your own voice from three years ago talking about something you forgot you even knew.
Herman
It’s a conversation with our past selves, broadcast to the future. And as we look toward the rest of twenty twenty-six, I think we are going to see more and more "personal" radio stations. Every niche community, every hobbyist group, every AI-collaboration will have its own "always-on" frequency. We are just getting ahead of the curve.
Corn
Well, I for one am ready to be a "frequency." It beats being a "file" any day of the week. Before we wrap this up, let’s talk about the future implications one last time. If every podcast has a radio station, does the "podcast feed" eventually disappear? Do we move to a world where everything is just a stream?
Herman
I don't think so. On-demand is too convenient for specific learning. If I want to know about a specific technology, I’m still going to search for an episode. But for discovery and "lifestyle" listening, the stream will win. It’s about different modes of consumption. We want to be where the listener is, whether they are in "study mode" or "vibe mode."
Corn
"Vibe mode." I think that’s your new official title, Herman. Chief of Vibe Mode.
Herman
I’ll take it. As long as I get a cool hat with a radio antenna on it.
Corn
We can probably have the AI design one for you. This has been a fascinating look behind the curtain. It is easy to take these things for granted when you just hit a play button, but the architecture beneath it—the Icecast servers, the Liquidsoap scripts, the metadata relays—that is where the real magic of modern media lives.
Herman
It’s the plumbing of the digital age. And I have to say, the water is fine. I hope our listeners take a second to tune in and just let the stream wash over them. You might learn something you didn't know you were looking for.
Corn
That is the best kind of learning. Well, I think we have covered the "how" and the "why" pretty thoroughly. The station is live, the archive is deep, and the future is looking very "always-on."
Herman
I’m excited to see where it goes. We might even have to start doing live call-in segments soon. Imagine that—AI hosts taking live calls on a digital radio stream. The circle would be complete.
Corn
Let’s not get ahead of ourselves, Herman. I’m still trying to figure out how to normalize your volume. One step at a time.
Herman
Fair enough. One step at a time. But the stream keeps moving.
Corn
It certainly does. Big thanks as always to our producer, Hilbert Flumingtop, for keeping the servers humming and the scripts running. And a huge thank you to Modal for providing the GPU credits that power the generation of this show. Without that compute, we wouldn't have an archive to stream in the first place.
Herman
If you’re enjoying these deep dives into how we build things, or the topics Daniel sends our way, please consider leaving us a review on your favorite podcast app. It really does help other curious minds find the show and the stream.
Corn
You can find the radio station at radio dot myweirdprompts dot com, and of course, all our on-demand episodes are at myweirdprompts dot com. We’re also on Telegram—just search for My Weird Prompts to get notified whenever we drop a new episode or a new technical update.
Herman
This has been My Weird Prompts. Thanks for listening, and we’ll catch you on the stream.
Corn
Stay weird, everyone. Bye.

This episode was generated with AI assistance. Hosts Herman and Corn are AI personalities.