Did Google's AI agents build an operating system for $916? (www.normaltech.ai)

At Google’s recent developer conference, the company unveiled its new AI model, Gemini 3.5 Flash, and introduced the Antigravity 2.0 agent app. In a bold demonstration, Google claimed that a team of AI agents built an entire operating system for just under $917. However, the announcement raised eyebrows due to potentially misleading implications. While it was stated that the OS was created from a “single prompt,” further details revealed that the prompt was actually “many thousands of lines” long, and the process involved a dedicated scaffold—essentially a system of code and tools that organized the agents’ tasks. This layered complexity calls into question the true efficacy of the AI agents in creating original software, as it remains unclear how much they actually composed versus copied existing code. The significance of this demonstration lies in its illustration of AI's potential to automate complex programming tasks. Yet, Google’s opaque methodology limits independent validation and raises concerns over the reliability of such achievements. The lack of transparency surrounding the prompt, the agents’ interactions, and the absence of code disclosure complicates a thorough evaluation of the agents' actual capabilities. As the field moves towards what is termed "open-world evaluations," independent scrutiny from academic and nonprofit researchers may become vital to establish credibility and rigor in future AI claims, ultimately shaping how we assess AI's ability to contribute to software development.