Wes McKinney explores the arithmetic shortcomings of large language models (LLMs) like Anthropic's Claude Code. He shares his experiences using these coding agents, highlighting how they can improve productivity but often struggle with basic calculations and reliability. Testing various models, he finds that local models perform better than many API options in handling arithmetic tasks.

The article discusses practical lessons for effectively working with large language models (LLMs), emphasizing the importance of understanding their limitations and capabilities. It provides insights into optimizing interactions with LLMs to enhance their utility in various applications.

The author shares insights from a month of experimenting with AI tools for software development, highlighting the limitations of large language models (LLMs) in producing production-ready code and their dependency on well-structured codebases. They discuss the challenges of integrating LLMs into workflows, the instability of AI products, and their mixed results across programming languages, emphasizing that while LLMs can aid in standard tasks, they struggle with unique or complex requirements.