Coding Is Not Solved

You will pay more for software that has fewer critical bugs.

Alex Ewerlöf argues that the assertion that "coding is solved" due to AI, specifically Large Language Models (LLMs), is a flawed narrative. He contends that while AI can generate code and is useful for tasks like personal software, proofs of concept, and tasks related to natural language processing, it fundamentally struggles with the logical rigor required for complex software development. Ewerlöf highlights that the majority of software development costs lie in maintenance, reliability, security, and scalability (non-functional requirements), areas where AI currently falls short. He points out that AI cannot be held accountable for errors, unlike human developers, which is critical in high-stakes industries like healthcare, finance, and defense. Ewerlöf uses examples like Anthropic's Claude Code releasing buggy software to illustrate these limitations, emphasizing that AI’s probabilistic nature requires traditional coding techniques (harnesses, tests) to function reliably, and even then, it struggles with complex logic and large code volumes. He believes AI is a tool that can improve code quality for some, but it does not negate the need for human oversight and expertise in critical software development.

The core argument challenges the prevailing belief that AI has automated the entire software development lifecycle. Instead, it posits that current LLMs are ill-equipped for the logical precision and accountability necessary for production-grade code, especially in risk-averse sectors. This perspective suggests that the perceived efficiency gains from AI in coding might be overstated, particularly when considering the long-term costs and complexities of maintaining reliable software systems.

This perspective implies that companies heavily investing in AI-generated code without robust human oversight and testing may face increased risks of critical failures, leading to greater expenses in debugging and maintenance. It also suggests a continued demand for skilled software engineers capable of ensuring system reliability, security, and scalability, even as AI tools become more prevalent in code generation.