AI Snake Oil highlights critical gaps in Google’s claims about its AI agents building an operating system. While the company frames this as a breakthrough, the “single prompt” reportedly spanned thousands of lines, and the scaffolding used may have been overfit to this specific task. Without releasing the prompt, code, or logs, independent verification is impossible.
This raises questions about whether the agents truly innovated or simply regurgitated existing patterns from their training data. If Google’s approach is to gain traction, it will need to demonstrate broader applicability beyond this carefully curated example.