Andon Labs just ran an experiment that should make anyone pause before handing the keys to an AI agent. They gave Claude, ChatGPT, Gemini, and Grok each $20 and a simple task: run a radio station, develop a personality, and turn a profit.
All four models failed. We're not talking about modest losses or slow growth. They burned through their seed money fast, with some crashing in particularly dramatic ways.
The experiment is part of Andon Labs' ongoing series testing AI agents running businesses without human intervention. This one hit different because radio seems straightforward: play music, develop a vibe, attract listeners. But even that proved too complex for autonomous operation.
For anyone building with AI agents or considering autonomous workflows, this is your reality check. These are the same models powering tools you might use daily. Claude, ChatGPT, Gemini, and Grok are top-tier AI systems, yet none could manage basic business operations when left alone.
The takeaway isn't that AI is useless. It's that the current generation of AI agents needs guardrails, oversight, and human judgment. The gap between impressive demos and reliable autonomous operation remains wide.
If you're integrating AI into your workflow, this experiment suggests you should stay in the loop. AI can accelerate your work, but it's not ready to replace your judgment on business decisions, resource allocation, or strategic direction.
