Why I will not be using LLMs in my coding | Kristian Polso

Software developer Kristian Polso said he will stop using large language models and AI agents in his professional coding work, citing cognitive costs and uneven output quality. Polso, who has built production machine-learning systems for about a decade, pointed to Microsoft research on self-reported reductions in cognitive effort among knowledge workers and to studies on “cognitive debt” and growing reliance on AI assistants. He said hours of LLM use leave him disoriented, and argued that verbose, heavily commented Claude-generated code can appear rigorous while containing bugs that fall to human reviewers to catch. He also echoed Esther Weidauer’s essay in framing LLMs not as a neutral tool but as a service that outsources work to a third-party company, typically US-based. The stance matters as developers debate whether AI coding assistants deliver durable productivity gains or quietly erode skills and oversight. What to watch next is whether employers and engineering teams measure those cognitive and review costs, or keep expanding LLM adoption on perceived speed alone.