News Q&A

What Did George Hotz Say About AI Coding Agents and Code Quality?

Historical record This story connects to a documented entry in the Agentic History timeline. Since high-profile autonomous coding-agent launches, the key question has shifted from capability demos to whether organizations can maintain quality and accountability at scale.

Answer: Decrypt reported hacker George Hotz published a blog post arguing that mass adoption of AI coding agents will be a costly mistake for software development. Hotz wrote that agents “cannot program” and produce broken output that is increasingly hard to detect, describing a dynamic where agents frontload apparent progress but leave uncertain finishing work.

The article said Hotz based his view on six months using agents on real projects, including parts of Tinygrad and firmware reverse-engineering of a USB-PCIe chip. It also contrasted his position with Andrej Karpathy’s recent pro-agent stance and noted claims that some Anthropic engineers rely on models for code while reviewing outputs.

Source: Decrypt (via Yahoo Tech)


← All news · Timeline · Research blog