Simple Made Inevitable: The Economics of Language Choice in the LLM Era
21 hours ago
- #Programming Languages
- #LLM
- #Clojure
- The article discusses the shift in language choice criteria from human convenience to machine efficiency in the LLM era.
- Clojure is highlighted for its simplicity, immutability, and token efficiency, which are beneficial for LLM coding agents.
- The 'brownfield barrier' is introduced as a problem where codebases become too complex for both humans and LLMs to navigate effectively.
- Token efficiency in Clojure allows for more context in LLM operations, giving it an advantage over more verbose languages like Python, JavaScript, and Java.
- Immutability in Clojure reduces the generation of defensive boilerplate code, a common issue with LLM-generated code.
- The REPL in Clojure provides a tight feedback loop, beneficial for iterative development with LLMs.
- The article contrasts 'easy' languages (familiar, comfortable) with 'simple' languages (objectively unentangled), advocating for the latter in the LLM era.
- Stability in Clojure's semantics over 17 years provides consistent training data for LLMs, unlike languages with frequent breaking changes.
- The investment question is framed around short-term vs. long-term language choice, with Clojure being favorable for long-term maintainability.
- Potential future languages designed specifically for machine cognition are hinted at, though current choices like Clojure already align well with machine needs.