Famed iPhone, Sony Hacker Says AI Coding Agents Are a Disaster Waiting to Happen

The software industry is facing a fundamental split regarding the utility of autonomous agents. George Hotz, the engineer who cracked the iPhone at age 17 and reverse-engineered the PlayStation 3, argues that the mass adoption of AI coding agents will be one of the most costly mistakes in the field’s history.

Hotz’s position is built on the observation that while high-performing engineers can identify flawed agent output, weaker engineers cannot. This creates a cycle where lower-quality code is produced at a higher volume, degrading the average quality of software at scale. He contends that agents cannot program and that their output is broken in ways that become increasingly difficult to detect as statistical models improve.

This debate is not merely academic. The tension is highlighted by the recent movement of top-tier talent. On May 19, 2026, Andrej Karpathy announced he had joined Anthropic's pre-training team, expressing a view that AI agents have already transformed software development. Hotz and Karpathy now represent opposite poles of a critical industry disagreement.

Hotz’s skepticism is grounded in direct application. He spent time using agents on real projects, including parts of Tinygrad and the reverse-engineering of a USB-PCIe chip. He describes the current state of agentic workflows as a "slot machine lever"—the agent provides initial progress, but the developer is left hoping the finishing work is completed.

The risk here is the accumulation of technical debt that is invisible to the uninitiated. If the industry moves toward a model where the volume of code increases but the ability to verify its integrity decreases, the long-term cost will be structural.

The question for engineering leaders is whether they are optimizing for the speed of initial output or the long-term stability of their codebase.