Currently, not many people are aware; I come from a technical background. Twenty years ago, at an investment bank in the Asia-Pacific region, there were probably about fifty people who truly understood algorithmic trading, and I was one of them. At that time, the core of the system was not just "functionality," but reliability over the long term under constraints such as latency, stability, monitoring, and risk management. Therefore, I was initially very opposed to "Vibe Coding," because it could easily lead people to misunderstand that programming could be done by "giving instructions based on feelings." Later, I personally used Claude Code, Codex, and Copilot to create some small software and internal tools, and I realized that the issue was not with AI writing code, but with the misleading nature of this term: when AI is integrated into the engineering process, it is more like an extension of Pair Programming, combining "production" and "review" into a single cycle, allowing people to shift their focus from repetitive work back to key decisions and quality control. Meanwhile, the reality is also very clear: people who do not understand programming will find it difficult to achieve satisfactory results with just Vibe; but for those who are capable, AI will significantly enhance efficiency and quality of delivery, which also explains why one of the most sought-after "advantages" in companies is having more LLM Tokens that can be used, controlled, and audited by everyone, including more stable access and model management, better permission and compliance frameworks, closer internal toolchains aligned with business needs, as well as ongoing training and the collection of best practices.