He started as a web front-end developer, quickly moved to back-end, and eventually found his niche in finance, bookkeeping, and payment processing. For years, he accumulated deep domain knowledge — how payment systems should be designed, what implementation trade-offs matter, how to prevent double billing in distributed transactions. This expertise was his moat. It's what made him valuable. It's what he thought AI could never touch.
In 2025, he joined a financial company that actively embraced AI. From day one, he was given ChatGPT and Claude Enterprise accounts and encouraged to use them. At first, he resisted. He valued his own knowledge and believed an LLM could never replace it. He created detailed design documents before coding, minimizing AI use. Then his supervisor pulled him aside: "Your code delivery pace is good, but you're taking too long to deliver the design documents. Are you using AI? You should be using more AI."
So he started using it. And that's when the first shock hit.
"I began to realize that the expertise I had accumulated over many years — regarding payment system design principles, implementation trade-offs, and designs to prevent double billing — could be recalled through prompt input to the AI. Although the model still needed adjustments, it was able to properly organize the most difficult parts that I had only just acquired in my mind through years of practical experience. That was my first real shock."
Next, he thought debugging was safe. Analyzing faults in distributed systems, investigating race conditions — that required years of experience. This ability was the key to his long-term employment. But with AI coding tools like Claude Code and GPT-5-Codex, bugs that previously took him a full day were now solved in a single step by AI.
"Of course, there's still a reason for me to be employed. People are needed to review code. But now I'm just another replaceable engineer. I don't have any expertise that other senior engineers piloting LLMs don't possess. My expertise in finance and payments, my debugging intuition honed through hours of sweat and tears, my knowledge of distributed systems — these are now things that can be pulled out at the prompt."
He's still working — for now. His company has resumed hiring for several positions, but specialized knowledge is no longer a major differentiating factor. "It's good for talented engineers who haven't had the opportunity to delve deeply into a specialized field — more job opportunities. But it's also sad that other talented engineers who have dedicated their lives to accumulating specialized knowledge now have to compete on the same playing field."
His blog post, published anonymously on a personal site, went viral across developer communities. On Hacker News, hundreds of comments poured in. Some agreed: "This is the reality, and it's always been that way in this industry. It takes about 10 years to realize it." Others pushed back, arguing that specialized domains like finance still require deep human oversight because AI agents "make mistakes regularly." But the author's response was measured and bleak: "I'm becoming an 'AI-native engineer,' reviewing and correcting AI-generated code. But this is merely the current optimal solution — and even this role may become unnecessary in the future."