On coding with LLMs
Summary
The article provides a cautious perspective on the current AI hype around LLMs, acknowledging real-use cases like quick scripting and code translation while warning that AI-generated code is not a finished product. It discusses limitations, references concepts like Amdahl's law and metaprogramming, and advocates for privacy-conscious, local AI use and the potential for abstractions or DSLs to manage AI-generated code. It also reflects on the author's consulting experience and suggests a path toward privacy-preserving features in tools like Aba Search and Replace.