George Hotz, the hacker who first cracked the iPhone at age 17 and reverse-engineered the PlayStation 3, published a blog post Sunday arguing that mass adoption of AI coding agents will result in significant degradation of software quality. Hotz wrote: "I'm calling it now, the adoption of AI agents into software development will be one of the most costly mistakes in the field's history." His position directly contrasts with Andrej Karpathy, one of AI's most prominent researchers, who joined Anthropic's pre-training team on May 19, 2026, explicitly stating his view that AI agents have already transformed software development. The two represent opposite poles of an unsettled industry debate, both with substantial credibility to support their positions.
Hotz's Core Argument
Hotz spent six months testing AI agents on real projects: parts of Tinygrad, his open-source deep learning framework, and a complete firmware reverse-engineering of a USB-PCIe chip. Based on this experience, he argues that "agents cannot program, and it's taking longer and longer to realize that they can't." He describes the output as "broken, but in a way that's getting harder and harder to detect. Which is exactly what you'd expect from an increasingly accurate statistical model."
His central concern is organizational. High performers have tight feedback loops to catch agent-generated problems before they ship—they read the code, spot errors, and calibrate when to trust the tool. "The bottom performers won't have that self check," Hotz writes. The critical issue: low performers are using agents to produce 10 times their previous output. At large companies, this produces faster degradation of average code quality, masked by sheer volume. Hotz describes the outcome as "a golden era for buckets and buckets of slop, and a dark age for gems of quality." He points to reports that Apple is pushing AI coding tools across its entire engineering organization, asking: "Do you think macOS will get better or worse in the next 2 years?"
Not About Ego
Hotz anticipates the objection that a programmer defining part of his identity through his craft would naturally resist replacement tools. He takes this seriously and dismisses it on merit. "Google's AFL found more bugs than LLMs and nobody felt that way about it. Chess and Go are more popular than ever," he wrote. Chess AI has dominated humans for decades while the game only grew more popular.
Hotz also expresses skepticism about industry motivations: "I almost think this is some kind of psyop to sell agents. Fear of loss is one of the only ways to make big companies move. Though I think in that fear they are making a big mistake."
Where the Camps Stand
Hotz now positions himself in what he calls the "LeCun/Marcus camp"—referring to Yann LeCun, Meta's chief AI scientist, and Gary Marcus, a longtime LLM skeptic. Both have argued that language models are fundamentally sophisticated pattern-matchers: they can imitate the distribution of existing code but cannot reason through genuinely new problems from first principles.
Vibe coding—describing what you want in plain language and letting AI generate the implementation—has exploded over the past year. Major AI labs have positioned agent-based coding as a flagship product. Microsoft transformed GitHub Copilot into a full agentic system in 2025, with CEO Satya Nadella describing it as a platform-level shift comparable to the move to cloud.
Karpathy had been skeptical of agents earlier in 2025 but reversed his position after new model releases. Anthropic CEO Dario Amodei stated in Davos that some Anthropic engineers have already stopped writing code themselves, letting models handle it while they review the output. Hotz, having tried the same approach, says he found himself reaching for the manual fix every time.
