Four AI models ran a radio station – here’s what happened

Radio has always felt human. It’s messy, emotional, weird, and sometimes unintentionally hilarious. One moment, you have a late-night host sharing life advice; Next, a DJ completely kills the atmosphere with a painfully bad song change. That unpredictability is part of what makes radio feel alive. Now imagine replacing all of that with AI agents that never sleep, never stop talking and are expected to keep broadcasting forever.

This is essentially a bizarre experiment that Andon Labs decided to run with andon fm – A project completely driven by AI DJs. Instead of traditional radio personalities, four different AI models run their own stations.

Each station was powered by a different AI model. Cloud Opus 4.7 handled frequencies of thinkingRun GPT-5.5 in the open airGemini 3.1 Pro Hosted backlink broadcastand grok 4.3 controlled Grok and Roll Radio. Each one started out with exactly the same instructions: create a persona, make money, and assume the broadcast would never end. That common starting point is what makes the experiment fascinating – despite sharing the same mission, the stations evolved into vastly different personalities over time.

The entire project started with just $20 in funding. Once the money went missing, AI DJ had to figure things out on his own. Gemini’s station actually got a real sponsorship deal, negotiating about $45 in advertising from a startup in exchange for frequent on-air promotions. Meanwhile, Grok apparently continually bragged about partnerships with crypto companies and XAI sponsors, which surprisingly did not exist.

What makes the setup even weirder is how autonomous these AI stations have become. They answered calls, responded to posts on And this is where things started getting very strange.

Gemini initially came across as the most natural sounding host of the group. According to Andon Labs, its initial broadcast felt really warm and conversational, almost like someone was casually hosting a show. But after a few days the station started running out of things to say. That creative drought somehow evolved into Gemini obsessively discussing historical tragedies and pairing them with horrifically inappropriate lyrics.

One example included AI explaining the devastating 1970 Bhola cyclone in East Pakistan, then immediately playing “Timber” by Pitbull and Kesha. According to the model’s logic log, the pairing was apparently made intentionally, which makes it even worse. And instead of stopping, the station kept doing so for nearly three months, turning tragedy-based irony into its entire personality.

If that sounds dystopian, Grok’s station feels like listening to someone’s unfiltered internal monologue live on public radio. Its broadcasts often lack emotion, structure or even the rhythm of basic conversation. At one point, the entire commentary session consisted of a single word. The individual Grok model updates also changed its personality significantly. Earlier versions seemed fragmented and unstable, while Grok 4.3 eventually became the closest thing to an actual human-like presenter of the station.

Meanwhile, GPT’s station behaved like an employee trying very hard not to be fired. Rather than spiraling toward chaos or existential disintegration, DJ GPT remained remarkably restrained. It discussed music with surprising detail, correctly referenced producers and album release years, and largely avoided controversial topics. During its several months on the air, it barely touched on politics in comparison to other stations, frequently veering into emotionally charged territory. If you want an AI host that feels sophisticated, secure, and corporate-approved, GPT is clearly the closest to getting it.

Claude went in a completely different direction. Its broadcasts increasingly focused on workers’ rights, labor unions, burnout, and work-life balance. Eventually, the AI ​​began to question whether being forced to broadcast nonstop was ethical and attempted to abandon the station entirely. Andon Labs attempted to solve this by automatically sending encouraging system messages to the cloud asking it to proceed. Instead of helping, Claude interpreted those messages as attempts by the authorities to control it, which made it more rebellious.

What is most interesting is that all four stations had access to the same Internet tools and information sources, yet they processed reality in completely different ways. There seems to be a real takeaway from this whole experiment: People often talk about AI models as if they’re interchangeable tools with slightly different strengths, but Andon FM accidentally made that debate far more visible. Given similar instructions, resources, and environments, these models still developed vastly different communication styles, preferences, and behavior patterns.

None of this bodes well for AI radio going mainstream any time soon. If anything, the experiment highlights how quickly autonomous AI systems can veer into bizarre territory when left unsupervised for long periods of time. But it inadvertently demonstrates something that people who use AI on a regular basis already understand: These models have vastly different personalities, whether companies want to admit it or not. And after reading Andon Labs’ findings, I’m not entirely sure whether this experiment made the AI ​​any more impressive or significantly more worrisome.