Despite being four years into the AI 'revolution,' I still do not use LLMs for anything I truly care about. The industry has little to show for it—software quality hasn't improved, it's not faster or cheaper to produce, and security remains unchanged.

A genuine technological revolution typically sees old systems replaced by new ideas, but AI boosterism has produced few innovations beyond navel-gazing. Everyone is shouting, but nobody is shipping useful products. I don't feel behind by avoiding LLMs; no one has displaced my open-source projects despite the noise.

Evidence for AI productivity gains is thin. $1.5 trillion later, there are almost no independent studies proving real-world improvements. Existing studies either measure irrelevant metrics or are too small to matter. The LLM-generated pull requests I receive are still poor quality—less repetitive but no more useful.

Even frontier models miss obvious issues that hobbyists catch with basic reasoning. The philosophy of AI boosters ignores decades of software theory. Code is an input, not an output. Peter Naur remains undefeated.

The quiet goal of big tech's LLM push is to homogenize intellectual labor, making it fungible and reducing worker bargaining power. By not adopting LLMs, I've carved out space to specialize in a smaller skill set, which is working well for my job security.