Why I remain a skeptic

I remain a skeptic regarding the widespread adoption of Large Language Models (LLMs) in software development. Despite the hype and claims, there is still no concrete evidence that LLMs have significantly improved the quality, speed, cost, or security of software. While there have been some advancements in certain areas, the industry as a whole has not seen a meaningful shift in capabilities or productivity.

In fact, most software remains the same and security concerns persist. The industry's focus on AI boosterism has led to a lack of tangible progress and innovation. Furthermore, there is a lack of independent studies to validate the productivity gains associated with AI. The few studies that exist often have limitations and suggest only marginal or even negative productivity improvements.

The PRs generated by LLMs are still of low quality, with obvious mistakes and inconsistencies. Although AI outputs may become slightly less formulaic, they still miss important details and cannot replace the human touch in software development. The philosophy of software development, which views code as an input rather than an output, has been challenged by AI boosters.

However, this approach contradicts the understanding that code is a crucial part of the development process. The push by big tech to homogenize intellectual labor through LLMs threatens the bargaining power of software developers. By not using LLMs, I have chosen to specialize in a narrower set of skills, which has improved my job security.

Overall, the evidence remains tenuous and thin, and my skepticism about the effectiveness of LLMs in software development remains strong.