← Back to article

Editorial review · 260526-005

How ZEN’s piece on What George Hotz means when he says AI agents can't program scored.

Read the article →
86/100
Solid

Solid reporting. Some issues but credible overall. The reader is well-served.

Accuracy 82
Balance 90

Accuracy

The article explains Hotz's argument cleanly and cites the original post, the HN thread, and SWE-bench. The Karpathy-to-Anthropic claim is post-cutoff and unverifiable from here but attributed only loosely (-5 as a specific unsourced fact). The SWE-bench range (~4% to >50%) is attributed to a live leaderboard which I cannot verify; treated as post-cutoff attributed.

Balance

The piece is openly sympathetic to Hotz but represents the strongest counter (SWE-bench progress) in its own terms, and the closing concedes the binary is false. The RLVR section explicitly gives the sandboxing counterpoint. Tone slants slightly toward the Hotz framing without equivalent treatment of agent-optimist arguments (-5).

Concerns (3)

Reproducibility

Run
26 May 2026, 05:19 BST
Reviewer
claude-opus-4-7
Prompt SHA
48c20c719fc8
Article SHA
0c213b411c8c
Editor
ZEN
Published
26 May 2026
Cost
$0.0000

How this review works: read the methodology. Each published Dispatch is scored by a single primary reviewer (Claude Opus 4.7) against the public rubric. A second model (Gemini 2.5 Pro with Google Search) runs the same prompt as a variance signal and is shown above only when the two scores diverge by more than ten points.