AI speeds code generation, but engineering accountability remains, author argues
Alex Ewerlöf says AI coding tools can help with prototypes and other lower risk work, while production software still demands human oversight.
AI tools can generate code faster, but they have not removed the need for engineers to understand, verify and take responsibility for software, argues Alex Ewerlöf in an article published on 28 September.
In “Coding Is Not Solved”, published on Ewerlöf’s blog and shared by Hacker News, he says AI can be useful for prototypes, personal software and some language-processing tasks. He cautions that professional systems often carry greater reliability, security and maintenance requirements.
Ewerlöf points to inconsistent model output and the difficulty of checking large volumes of generated code. In his view, testing and other safeguards can reduce errors, but do not replace an engineer’s understanding of how a system works or the judgement needed to fix it when it fails.
The article frames accountability as a central constraint: people remain responsible for software they ship, even when AI helped produce it. Ewerlöf argues that this matters most where mistakes could carry financial, legal or safety consequences.
He also suggests engineers’ work could shift towards product development, AI deployment and quality roles. The article offers his views and experience; it does not quantify AI’s effects on software quality, costs or employment.
