AI coding agents — George Hotz’s ‘The Eternal Sloptember’
AI coding agents feature in George Hotz’s blog post titled ‘The Eternal Sloptember,’ in which he critiques their use in software development. Hotz says he spent six months testing agents on real projects. He warns of an avalanche of undetectable slop and a decline in code quality at scale, and the article notes that Andrej Karpathy joined Anthropic’s pre-training team five days before the blog post’s publication.
The Eternal Sloptember
In his blog post ‘The Eternal Sloptember,’ George Hotz presents several key criticisms of AI coding agents. Hotz argues that these agents are incapable of genuine programming, stating, “Agents cannot program, and it’s taking longer and longer to realize that they can’t.” He highlights the issue of their output being broken in ways that are increasingly difficult to detect, attributing this to their nature as “an increasingly accurate statistical model.”
Furthermore, Hotz uses the metaphor that AI agents “frontload all the progress,” leaving developers with a “slot machine lever” to pull, hoping for the completion of remaining tasks. He speculates that these agents might be pushed through fear tactics, suggesting, “Fear of loss is one of the only ways to make big companies move,” potentially leading to costly mistakes.
Additionally, Hotz draws comparisons with Google’s AFL bug-finding tool and popular games like Chess and Go, suggesting these AI systems perform poorly in contrast, with Google’s tool finding more bugs while not receiving the same reliance as AI coding agents.
Andrej Karpathy joined Anthropic’s pre-training team, with a May 19, 2026 date referenced in his post. The article reports that Karpathy’s announcement occurred five days before George Hotz published his blog post ‘The Eternal Sloptember.’ The article links Karpathy’s May 19, 2026 post to the timeline surrounding Hotz’s critique of AI coding agents. The article does not provide any additional details about Karpathy’s role beyond stating his placement on Anthropic’s pre-training team. These timing details are presented alongside reporting on Hotz’s views of AI coding agents.
The Eternal Sloptember
George Hotz published a blog post titled ‘The Eternal Sloptember‘ that critiqued AI coding agents after spending six months testing agents on real projects. In the post he warned of an avalanche of undetectable slop and a decline in code quality at scale.
The article notes that Andrej Karpathy joined Anthropic’s pre-training team five days before the blog post was published.
