This blog post discusses the reliability of autonomous agents compared to humans over extended workflows. It highlights that no autonomous agents have created complex, reliable software, despite their equivalent untrustworthiness at a per-step level. The author raises questions about why humans can achieve what autonomous agents cannot.