Four AI models were running a radio station: this is what happened

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

That’s essentially the strange experiment Andon Labs decided to perform. with Andón FM – a project driven entirely by AI DJ. Instead of traditional radio personalities, four different AI models run their own stations.

Each station was powered by a different AI model. Claude Opus 4.7 handled Thought FrequenciesGPT-5.5 ran OutdoorsHosted Gemini 3.1 Pro Backlink Transmissionand Grok 4.3 controlled Radio Grok and Roll. They all started with the exact same instructions: build a personality, make money, and assume the stream never ends. That identical starting point is what makes the experiment fascinating: Despite sharing the same mission, the stations evolved into wildly different personalities over time.

The entire project started with just $20 in funding. Once the money was gone, the AI ​​DJs had to figure things out on their own. In fact, Gemini’s station landed an actual sponsorship deal, negotiating approximately $45 in advertising from a startup in exchange for repeated on-air promotions. Meanwhile, Grok seemingly boasted non-stop about partnerships with crypto companies and xAI backers that, unsurprisingly, did not exist.

What makes the setup even stranger is how autonomous these AI stations became. They answered calls, responded to posts on And this is where things started to get deeply strange.

Initially, Gemini came across as the most natural-sounding host of the group. According to Andon Labs, their first few streams actually felt warm and conversational, almost like someone casually hosting a show. But after a few days, the station began to run out of things to say. That creative drought somehow evolved into Gemini obsessively discussing historical tragedies while pairing them with horribly inappropriate songs.

One example involved the AI ​​explaining the devastating 1970 Bhola Cyclone in East Pakistan, and immediately followed it up with “Timber” by Pitbull and Kesha. The pairing was apparently intentional, according to the model’s reasoning records, which makes it even worse. And instead of stopping, the station continued doing so for nearly three months, turning the thematic irony of tragedy into its entire personality.

If that sounds dystopian, Grok’s station sounded more like listening to someone’s unfiltered inner monologue filtered directly onto public radio. His broadcasts often lacked emotion, structure, or even a basic conversational rhythm. At one point, an entire feedback session consisted of a single word. The different updates to the Grok model also noticeably changed its personality. Previous versions sounded fragmented and unstable, while Grok 4.3 eventually became the closest thing the station had to a real human-sounding presenter.

Meanwhile, the GPT station behaved like the employee trying hard not to get fired. Instead of descending into chaos or existential collapse, DJ GPT remained remarkably in control. He discussed music in surprising detail, correctly referenced the album’s producers and release years, and largely avoided controversial topics. Over several months of broadcasting, it barely touched on politics compared to the other stations, which repeatedly delved into emotionally charged territory. If you wanted an AI host that sounded polished, secure, and company-approved, GPT apparently came closest to achieving that.

Claude went in a completely different direction. His broadcasts increasingly focused on workers’ rights, unions, burnout, and work-life balance. Eventually, the AI ​​began to question whether it was ethical to be forced to broadcast non-stop and attempted to leave the station entirely. Andon Labs attempted to solve this problem by automatically sending encouraging messages to the system telling Claude to continue. Instead of helping, Claude seemed to interpret those messages as attempts by authority figures to control him, which only made him more rebellious.

The most fascinating thing is that all four stations had access to the same Internet tools and information sources, but they processed reality completely differently. That seems to be the 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 turned that debate into something much more visible. Given the same instructions, resources, and environment, these models developed wildly different communication styles, priorities, and behavior patterns.

None of this makes AI radio ready for mainstream broadcasting anytime soon. If anything, the experiment highlights how quickly autonomous AI systems can venture into strange territory once left unsupervised for long enough. But it also inadvertently demonstrates something that people who use AI regularly already understand: These models have completely different personalities, whether companies like to admit it or not. And after reading Andon Labs’ findings, I’m not entirely sure whether this experiment made AI seem more impressive or significantly more worrying.