AUTO-UPDATED

Coding Is Not Solved – Alex Ewerlöf Notes

Software engineering remains a complex discipline requiring human accountability, as current AI models struggle with logic, reliability, and the non-functional requirements essential for production-grade software systems.

Key Points

  • AI models are stochastic and lack the accountability required for critical sectors like healthcare, finance, and defense.
  • Coding is not a "solved" problem because maintenance, security, and scalability represent the majority of software development costs.
  • LLMs often exhibit "jagged intelligence," making them inconsistent and difficult to debug compared to deterministic code.
  • Over-reliance on AI for code generation can lead to "AI overdose," where developers lose the ability to reason about system behavior.
  • Effective engineering requires deep understanding and ownership, which cannot be fully offloaded to automated tools without risking service degradation.

Why it Matters

The push to replace human engineering with AI-generated code risks prioritizing short-term velocity over long-term system reliability and security. Businesses that fail to maintain human oversight and accountability may face significant technical debt and operational failures when automated systems inevitably encounter logical errors.
Alexewerlof.com Published by Alex Ewerlöf
Read original