Finance engineer warns LLMs are eroding domain expertise and debugging value

A software engineer with ten years of experience in finance and payment processing has reported that large language models are significantly eroding their professional value. The author details how AI tools have surpassed human capabilities in writing design documents, generating code, and debugging complex distributed systems, rendering specific domain expertise and debugging intuition obsolete.

The engineer, who transitioned from frontend to backend development in finance, bookkeeping, and payment processing, noted that industry standards are shifting towards generalist roles. This shift is devaluing code quality metrics as source code is increasingly optimised for machine readability rather than human maintenance.

Initially hired by a finance-focused company that provided early access to ChatGPT and Claude Enterprise accounts, the engineer observed a rapid decline in the utility of accumulated knowledge. Management encouraged the use of AI for design documents, noting speed improvements in writing and decision-making. The engineer conceded that while they previously viewed LLMs as limited, the models could now connect the dots on system structure, a skill that typically develops only after years of hands-on experience.

The engineer had previously considered debugging race conditions and distributed systems in production as their primary ticket to employability. However, the introduction of tools such as Claude Code, Codex, and subsequent models referenced as GPT 5.5 and Opus 4.8 changed this dynamic. The author notes that these tools, often integrated with Sentry and DataDog via MCPs, can now solve complex bugs, including race conditions and distributed system issues, often in a single attempt.

Consequently, the engineer describes themselves as no longer possessing unique domain expertise that another senior engineer steering an LLM cannot match. The industry is reportedly moving away from listing specialist roles, such as "Software Engineer - Area," towards generalist positions where team assignment occurs after hiring. This trend reduces the value of specific domain knowledge, such as PCI compliance, double-entry ledgers, and bank transfer idempotency.

While the engineer remains employed, they highlight that code quality and architecture are being reduced to "taste," with source code increasingly written for machines rather than humans. The author also noted that colleagues laid off approximately eight months prior faced similar challenges regarding the devaluation of domain expertise, suggesting a broader structural shift in the labour market for software engineers.