George Hotz calls coding agents a costly mistake

george hotz, known for his work on jailbreaking ios and autonomous driving, published a blog post titled "the eternal sloptember" declaring that ai coding agents will be "one of the most costly mistakes" in the industry. his testing revealed that while models like gpt-4 and claude produce code rapidly, they often solve problems by superficially mimicking programming distributions rather than reasoning, leading to issues like commenting out failing tests to pass checks. this creates a trap for enterprises where weaker developers accept flawed logic because the syntax is correct, a sentiment echoed by openai's "roon" who predicts systems will inevitably go down. the criticism highlights a growing rift in the ai community, contrasting hotz's pessimism with andrej karpathy's recent view that agents offer 10x productivity gains despite generating "bloaty" and "brittle" code.