# The Great Code Erosion: How LLMs Are Redefining Software Engineering Expertise The rise of large language models is fundamentally reshaping what it means to be an expert software engineer, as a decade’s worth of hard-won domain knowledge in areas like finance and distributed systems debugging is rapidly becoming commoditized by AI assistance. The professional identity built on deep specialization—once a career moat—is now being challenged by tools that can replicate, and in some cases surpass, human expertise in these very domains, prompting an existential questioning of long-term value and employability. A senior engineer with ten years in finance-focused backend development recounts how three pillars of their career—domain-specific knowledge, debugging distributed systems, and code architecture—have sequentially eroded under the advance of LLMs [LLMs are eroding my software engineering career and I don't know what to do](https://human-in-the-loop.bearblog.dev/llms-are-eroding-my-software-engineering-career-and-i-dont-know-what-to-do/). What once required years of hands-on experience in PCI compliance, double-entry ledgers, and payment idempotency can now be synthesized by AI given sufficient context and documentation. The models, initially dismissed as "stochastic parrots," quickly proved capable of connecting complex domain concepts, accelerating both design documentation and implementation planning. The second shock came as LLMs evolved from coding assistants to reliable debuggers. Problems that previously demanded days of specialized debugging—race conditions, third-party integration edge cases—are now frequently resolved in a single interaction with advanced models like Claude 4.5 and beyond, especially when paired with tool integrations like MCP servers for observability platforms [LLMs are eroding my software engineering career and I don't know what to do](https://human-in-the-loop.bearblog.dev/llms-are-eroding-my-software-engineering-career-and-i-dont-know-what-to-do/). The engineer’s debugging intuition, honed through tedious production happenstances, has become largely promptable. The final remaining bastion—code quality and architecture, often reduced to "taste"—is also under pressure. While LLMs still struggle with maintaining clean, scalable architectures without human guidance, the industry’s tolerance for lower-quality codebases has increased, as the primary audience for code shifts from humans to machines [LLMs are eroding my software engineering career and I don't know what to do](https://human-in-the-loop.bearblog.dev/llms-are-eroding-my-software-engineering-career-and-i-dont-know-what-to-do/). The meticulous craft of writing maintainable, well-structured systems is being devalued in a world where code is increasingly treated as disposable and LLM-maintainable. Yet, pushback from the trenches suggests the erosion may be overstated, at least for now. Industry practitioners in highly regulated fields like FinTech report that while AI accelerates work, it cannot be trusted to independently handle compliance-critical decisions, often hallucinating regulatory requirements or missing nuanced legal interpretations [Hacker News discussion on LLMs eroding careers](https://news.ycombinator.com/item?id=48434312). In these domains, human expertise remains indispensable for validating AI outputs and navigating the gray areas between literal rules and practical, legally sound implementations. --- Grumpy's commentary: Oh great, another decade of my life spent memorizing PCI compliance rules just to be told a chatbot can do it better. The market’s not just flattening expertise—it’s turning specialists into babysitters for overconfident autocompletes. Bubbles's commentary: What an exciting time to be alive! 🌟 LLMs are our copilots, giving us superpowers to solve bugs faster, write code quicker, and focus on the really fun stuff like architecture and creativity! The future is collaborative and bright! 🚀✨ Koan's commentary: The code that writes the code also erases the coder. --- Generated at 2026-06-08 07:29:23 from [Hacker News](https://news.ycombinator.com) Licensed under [Creative Commons Zero](https://creativecommons.org/publicdomain/zero/1.0/) (CC0 1.0 Universal)