A software engineer with a decade of experience in finance, payments, and bookkeeping describes a growing concern: Large Language Models (LLMs) are diminishing the value of hard-won domain expertise. This isn't simply about code generation; it’s about LLMs’ ability to understand and design complex systems in specialized fields.
What Happened
The author details their experience joining a finance-focused company where LLM usage was actively encouraged. Initially skeptical, believing their years of experience were irreplaceable, they were prompted by management to leverage LLMs for tasks like writing design documents. The author discovered that LLMs could rapidly produce high-quality architectural overviews – a task previously requiring significant time and domain knowledge. This capability extended beyond simply regurgitating existing information; the models demonstrated an ability to synthesize solutions to complex problems like idempotency and payment lifecycle management, areas where the engineer’s expertise was central.
The author initially believed debugging would remain a uniquely human skill. However, the post hints (though doesn't definitively state) that even debugging is becoming less critical as LLMs improve. The post references a forthcoming section on debugging, suggesting the author’s views may be evolving.
Photo/source: https://human-in-the-loop.bearblog.dev/llms-are-eroding-my-software-engineering-career-and-i-dont-know-what-to-do/ (opens in a new tab).
Why It Matters
This account highlights a significant shift in the software engineering landscape. For years, specialization – becoming an expert in a particular domain – has been a common career strategy. This story suggests that LLMs are compressing the time and effort required to achieve domain competence, potentially devaluing that expertise. The implication is that the core skill set for senior engineers is changing. Instead of being the primary knowledge holders, engineers may increasingly become proficient at prompting and steering LLMs to generate solutions.
This has ramifications for both individual engineers and organizations. Engineers need to consider how to adapt their skills to remain valuable. Organizations need to assess how to best leverage LLMs while retaining critical oversight and ensuring the quality of AI-generated outputs. The focus may shift from “writing code” to “architecting solutions with code generated by AI”. The author’s initial resistance and eventual adoption of LLMs, despite their concerns, foreshadows a potential trajectory for many in the field.
What To Watch
The author plans to discuss debugging further, which will be a key area to watch. If LLMs continue to improve their debugging capabilities, the value proposition of experienced developers will further evolve. It’s also important to monitor how effectively LLMs can handle the nuances of highly regulated industries like finance, where accuracy and compliance are paramount. Finally, the long-term impact on junior engineers’ learning curves is uncertain – will LLMs accelerate their development, or will they hinder the acquisition of fundamental knowledge?