Coding Is Not Solved – Alex Ewerlöf Notes

Services you rely on may start experiencing more bugs as AI is prioritized over software quality.

Alex Ewerlöf argues that the common assertion that "coding is solved" by AI, specifically LLMs, is a flawed narrative. While acknowledging AI's utility as a tool for tasks like generating code snippets or proofs of concept, Ewerlöf emphasizes that the core complexities of software development, particularly non-functional requirements (NFRs) such as maintenance, reliability, security, and scalability, remain unsolved. He posits that LLMs struggle with the logical rigor required for complex systems, especially in high-stakes industries like healthcare, finance, and defense, where accountability is paramount. Ewerlöf points out that AI cannot be held accountable as it lacks the capacity for consequences, unlike human developers. He criticizes the over-reliance on AI, suggesting it can lead to degraded service quality and a misunderstanding of software engineering's true cost.

The claim that LLMs have "solved" coding overlooks the critical, high-cost areas of software development: maintenance, reliability, and security. Ewerlöf argues that AI's current limitations in logical reasoning and accountability make it unsuitable for mission-critical applications where failure has severe consequences. This suggests a market where the hype around AI-driven code generation may be outpacing its practical, safe application in robust systems, potentially leading to increased development debt and user-facing issues in the short to medium term.

This perspective signals a potential disconnect between AI tool vendors and the realities of production software engineering, particularly for large enterprises. Companies pushing AI integration without addressing fundamental NFR challenges risk customer dissatisfaction and reputational damage, as Ewerlöf highlights with examples of degraded services. Developers and leaders are urged to resist the pressure for AI adoption at the expense of essential quality and reliability, indicating a need for more nuanced AI integration strategies that prioritize long-term system health over short-term velocity.