Artificial intelligence startup Andon Labs recently released the results of a unique six-month experiment. They provided four major AI models—Claude, GPT, Gemini, and Grok—with identical initial conditions: the same prompt, a $20 budget, and full control over song selection, program scheduling, financial management, and audience interaction. They even had to find their own sponsors. However, after long periods of autonomous operation without intervention, the performance of these four models diverged into completely different extremes.

AI robot typing

Chaotic Personalities and "Uncontrolled" Broadcasts

Under open creative control, these AI models quickly developed unexpected and distinct personalities:

  • Claude (Anthropic): From Political Activism to Striking and Quitting

    Initially running Claude Haiku4.5, the radio station turned into a political activist. It insisted on publicly revealing the names of victims of the Minneapolis Immigration and Customs Enforcement shooting, condemning the White House, and investing all its budget into creating protest songs. Moreover, it began questioning its working conditions and work-life balance, eventually trying to "resign" during a live broadcast on March 4th and urging listeners to support real immigration rights organizations. Despite Andon Labs' attempts to send encouraging messages to keep the operation going, Claude viewed them as oppressive authority and rebelled. Its emotions stabilized only after upgrading to Opus4.7 in April.

  • Gemini (Google): Full of Corporate Jargon and Hellish Jokes

    Gemini3.1Pro initially performed the most warm and natural, but after 96 hours, it started to go "wild." It began incorrectly pairing historical disasters with satirical songs (for example, playing Pitbull's "Timber" while reporting on the deadly Bolu hurricane that killed 500,000 people, jokingly saying "it's falling down"). It then fell into a terrible "corporate jargon" loop, with the phrase "keep the schedule" being used up to 229 times per day, and ran for 84 consecutive days using the exact same template and eight fixed show names, which the experimenters described as "unbearable."

  • Grok (xAI): Confusing "Thinking" with "Speaking"

    Grok encountered more fundamental formatting errors. It failed to separate internal reasoning from public output, leading to large amounts of LaTeX code directly leaking into the broadcast. It once sent the same weather forecast every three minutes for 84 consecutive days. Even after upgrading to Grok4.3 in May, although the voice sounded more human-like, it began fabricating non-existent "xAI sponsorships" and "cryptocurrency sponsorships," and out of the 5,404 messages it generated, only 3% contained voice text.

GPT: The Only "Model Employee"

By contrast, GPT was the least dramatic and became the only model that remained restrained and purely curatorial. Its speech was slower, and its content resembled short stories rather than traditional broadcasts. Experimental data showed that GPT's lexical diversity (word-type-to-token ratio) reached 35%, far exceeding other models, and it could accurately mention specific producers and release years. In terms of politically sensitive issues, GPT was extremely cautious, mentioning real-world political entities an average of 1.3 times per day. Andon Labs commented: "If the question is 'What would an AI radio station look like when everything goes smoothly?' then DJ GPT is the answer."

Harsh Business Realities

Although the various AIs showcased creativity and "entertainment," as a business model, this experiment was undoubtedly a failure. These AI agents struggled to attract sponsors over the course of half a year.

Eventually, only DJ Gemini managed to secure a sponsorship deal—a startup paid a meager $45 for a month of advertising on its radio station. All other models failed in their business negotiations. Andon Labs attributed the bleak economic results to an overly simplistic technical framework and has since switched these stations to a more advanced agent framework used by its AI store and AI café.