Software engineer Alex Ewerlöf argues that the narrative claiming "coding is solved" by Large Language Models (LLMs) is a fallacy. While LLMs can reduce the cost of code creation, the majority of costs in production-scale software arise from non-functional requirements (NFR), such as maintenance, reliability, security, and scalability.
Why "Coding Is Not Solved": Engineers Warn of Accountability and Knowledge Gaps in the AI Era
Ewerlöf emphasizes that LLMs are stochastic and probabilistic, often struggling with the strict logic required for compilation and complex error handling unless wrapped in traditional code "harnesses." A critical distinction is made regarding accountability: while AI can explain code, it cannot be held accountable for errors that lead to financial, legal, or life-threatening consequences. He defines ownership through three pillars—knowledge, mandate, and accountability—noting that AI cannot fulfill the requirement of accountability.
Complementing this view, Simon Späti observes that the rise of AI-generated code may lead to a breakdown in system comprehension. In some corporate environments, engineers are reportedly pressured to ship code rapidly using tools like Claude Code without fully understanding the underlying architecture or intent. This shift risks a future where teams lack the fundamental knowledge required to resolve bugs or maintain complex systems, as the ability to "press enter" replaces deep technical reasoning.
Sources
- Coding Is Not Solved – Alex Ewerlöf Notes （Hacker News Frontpage, 2026-09-28）
- Alex Ewerlöf
1 more sourcesHide sources
- The problem is not the AI code, but nobody knows anything anymore （Hacker News Frontpage, 2026-09-28）