AI Coding Published on Sep 12, 2025 --- The author expresses skepticism about popular beliefs around AI coding, emphasizing a harsh reality about truth and perception in today's markets. Key Points Truth vs. Market Interest Most people prefer narratives that inflate financial value ("pump their bags") rather than seek actual truth. This mindset drove billions into failed self-driving car ventures and now fuels unrealistic expectations for AI coding. AI Coding Compared to Compilers The best analogy for programming AI is a compiler: You input a prompt (like code). The AI outputs "compiled" code. Interactive adjustments post-output often fail; better to revise the original prompt. However, unlike programming languages with strict specs, AI uses English: English is imprecise and verbose for complex or novel programming tasks. AI output is non-deterministic; the same prompt can yield varying results. Prompts are non-local; changes anywhere can unpredictably affect output. In short, AI coding "works" largely because current programming tools, languages, and libraries are flawed. AI as a Tool, Not a Replacement AI advancements may yield better tools through enhanced search, optimization, and pattern recognition. But the programmer still codes, just using a different "language." Code Quality and Hiring Standards Matter The ability of AI to replace developers reveals poor codebases or low hiring bars in some companies. Historical Context of Job Replacement AI will gradually replace programming jobs similarly to how compilers replaced some coding tasks and spreadsheets affected accounting jobs. A Call for Better Foundations Instead of chasing hyped AI solutions, focus should be on improving foundational elements: Programming languages Compilers Libraries --- Criticism of AI Hype The author condemns billions wasted on trendy tech with little real impact. References a study showing AI makes users feel 20% more productive but actually slows them by 19%. Warns against mistaking market hype for meaningful progress. --- Conclusion The future of programming involves viewing AI coding as an advanced tool—akin to a compiler—integrated thoughtfully into workflows. Real progress demands deeper investment in better software engineering foundations rather than chasing fancy narratives or hype-driven funding. --- Contact & About Author alias: the singularity is nearer Contact: geohot@gmail.com Social: GitHub, Twitter Blog mission: A home for poorly researched ideas often revisited by the author.