I'm done using AI

For the past two decades, the author has been immersed in programming, starting as a hobbyist and turning it into a rewarding career. Over the last 18 months, their profession, pastime, and passion have been drastically transformed by the arrival of language model-powered coding assistance (LLM-powered coding harnesses, such as Claude Code) and "agentic programming."

Due to its adverse effects on the environment, society, and mental well-being, the author has opted to cease using such tools. Initially, the author became acquainted with LLMs for programming through GitHub Copilot integration in Visual Studio Code, which provided a preview of potential implementations as the user typed. However, the author found this feature distracting and ultimately useless.

The author also encountered issues with LLMs providing incorrect solutions when colleagues sought help with debugging, as the models fabricated non-existent software library versions. As the author's organization adopted AI-driven coding tools, conversations shifted towards integration, even though the author continued to code traditionally, focusing on tests and features.

The author's colleagues utilized AI for coding, but the author remained content with their established workflow, finding no need to hasten the process. In 2025, management asked the author about their thoughts on AI coding tools, to which the author replied that they were often incorrect, hindered their coding speed, and did not enhance productivity.

After a year of experimenting with various AI-powered coding tools, the author discovered agentic editing, which made the experience feel magical. However, during mid-2026, the author learned that LLM-based coding harnesses could build a non-trivial project from start to finish without the author editing any code. While this may seem impressive, the author argues that there are significant drawbacks to relying on AI for coding, including the potential for the technology to make users delusional and even hazardous.