I will not be using LLMs in my coding

If you rely on AI for coding, your output may contain subtle bugs and become harder to maintain over time.

Kristian Polso, a developer based in Finland, has announced he will not be using Large Language Models (LLMs) or AI agents in his professional coding work. Polso cites several "dealbreakers" for his decision, including the cognitive impact such as "LLM brainfog" and disorientation, concerns about the quality and maintainability of LLM-generated code, and a personal preference for the process of writing code and understanding human-written logic. He also raises environmental concerns related to AI data centers and the political implications of relying on US-based AI companies, citing the CLOUD Act's implications for EU citizens. While acknowledging potential minor benefits for getting unstuck, Polso concludes that the drawbacks outweigh the advantages, opting to return to "old school coding."

Polso's rejection of LLMs highlights a growing concern among some developers regarding the "service" nature of AI coding assistants, contrasting it with the perception of them as mere tools. This perspective suggests that outsourcing code generation to third-party AI providers introduces dependencies and risks beyond the purely technical, touching on data privacy and vendor control. The argument is that these systems are not truly neutral tools but integrated services with inherent limitations and potential downsides that developers must actively consider.

The author's emphasis on the cognitive impact and the potential for LLMs to obscure code quality indicates a market trend where the ease of AI generation might inadvertently widen the gap between novice and expert developers. This raises questions about the long-term development of programming skills and the potential for AI to foster a generation of coders less attuned to code elegance and maintainability. Polso's stance suggests a pushback against the uncritical adoption of AI in development, advocating for a more mindful and human-centric approach to software creation.