A Hacker News discussion of an article titled "LLMs are eroding my software engineering career and I don't know what to do" reached 787 points and 767 comments, revealing widespread concern among software engineers about AI automation displacing traditional expertise. The post, written by a 10-year veteran engineer, sparked intense debate about the reliability of LLM-generated code and the future value of human domain knowledge.
LLM Hallucinations Create Critical Risks in Regulated Industries
Multiple engineers working in regulated domains reported dangerous reliability issues with advanced AI models. A FinTech worker described how an AI model "confidently asserted" regulatory non-compliance that was factually incorrect, having "hallucinated what the regulation actually required." This pattern of authoritative-sounding but incorrect outputs appears across domains.
Another commenter noted their "most capable agent...is regularly wrong, frequently myopic, and just outright dumb constantly," requiring constant human oversight to correct errors. The inconsistency problem extends to basic reproducibility: even with identical context, models produce different results, making reliability difficult to predict.
One engineer highlighted the code review burden: "can't ask an LLM how well they can code in Ada...I have to spend money and time code reviewing before formulating expectations." This represents a fundamental shift from direct productivity to oversight roles.
Human Expertise Remains Essential Despite Automation Gains
While LLMs demonstrate efficiency in handling routine engineering tasks, experienced engineers provide critical value in oversight and judgment. Discussion participants emphasized that expertise helps teams "push back on track" when AI agents make mistakes, particularly in complex domains requiring contextual understanding rather than literal rule-following.
The discussion revealed a paradox: LLMs are powerful enough to automate many tasks previously performed by senior engineers—domain expertise, debugging, architectural decisions—but unreliable enough to require expert validation. This creates uncertain career trajectories as the role of software engineers shifts from direct implementation to AI oversight.
Key Takeaways
- A Hacker News post about LLM career displacement reached 787 points with 767 comments, indicating widespread engineer concern
- AI models in regulated industries like FinTech have "confidently asserted" false regulatory interpretations, creating compliance risks
- LLMs produce inconsistent results even with identical context, requiring extensive human code review
- Engineers report their "most capable agent" remains "regularly wrong, frequently myopic, and just outright dumb constantly"
- Human expertise continues to provide essential value for error correction and judgment in complex domains where AI automation falls short