Software Engineering fundamentals matter more

LLMs excel at narrow coding tasks but cannot replace deep engineering reasoning like maintainability and architecture.

The author argues that despite crossing the "can it be done" threshold, LLMs still lack the reasoning needed for debuggable, layered, composable software. Open-weight models are increasingly viable. LLMs predict rather than reason, as shown in Apple's The Illusion of Thinking paper. The lethal trifecta (Simon Willison) highlights their inability to distinguish good advice from bad, making prompt injection a fundamental gap. The author advocates careful human judgment in designing abstractions and managing cognitive load, treating LLMs as collaborative tools rather than replacements.