Quoting Andrew Kelley

2026-04-30 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Andrew Kelley, creator of the Zig programming language, asserts that it is possible to detect the use of Large Language Models (LLMs) in code contributions, despite a common misconception. He notes that while not all LLM-assisted pull requests (PRs) are caught, the errors generated by LLMs fundamentally differ from human mistakes, making them identifiable. Kelley introduces the concept of a "digital smell" associated with agentic coding, which is apparent to those who do not use such tools, likening it to a non-smoker detecting a smoker's presence. This observation underscores a distinct stylistic or error-pattern difference between human and AI-generated code.

Key takeaway

For engineering managers evaluating code contributions, recognize that LLM-assisted code often carries a distinct "digital smell" and unique error patterns. Implement review processes that specifically look for these AI-generated characteristics to maintain code quality and ensure human accountability, especially in critical projects like the Zig language.

Key insights

LLM-generated code exhibits a detectable "digital smell" due to distinct error patterns and stylistic differences.

Principles

• LLM errors differ fundamentally from human errors.
• Agentic coding leaves a detectable stylistic signature.

In practice

• Review code for LLM-specific hallucination patterns.
• Identify "digital smell" in agentic coding contributions.

Topics

• LLM Detection
• Andrew Kelley
• Agentic Coding
• Digital Smell
• LLM Hallucinations

Best for: CTO, VP of Engineering/Data, Software Engineer, AI Engineer, Director of AI/ML

Related on AIssential

• The TechBeat: Can LLMs Generate Quality Code? A 40,000-Line Experiment (1/6/2026) — LLMs generate sloppy code over time, requiring formal analysis and multi-model reviews for quality assurance.
• Configuration Smells in AGENTS.md Files: Common Mistakes in Configuring Coding Agents — Configuration "smells" in AGENTS.md/CLAUDE.md files are widespread, increasing costs and reducing coding agent effectiveness.
• Every AI output is a hallucination — Large Language Models inherently produce fluent, confident outputs that are often wrong yet indistinguishable from truth, posing significant reliability issues.
• Maintainability sensors for coding agents — Maintainability sensors, combining computational and inferential tools, significantly enhance AI coding agent output quality and human review efficiency.
• Turning Andrej Karpathy’s LLM Coding Thoughts into Claude.md — LLMs are transforming coding workflows, shifting developers from direct coding to guiding and reviewing AI agents.
• Beyond Correctness: Enhancing Architectural Reasoning in Code LLMs via Scalable Labeling with Agentic Judgment — An agentic LLM pipeline scalably labels architectural quality, significantly improving code LLM reasoning and patch conformance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.