Two things LLM coding agents are still bad at
16 hours ago
- #LLM limitations
- #coding agents
- #human vs AI development
- LLMs don't use copy-paste or cut-paste functions when handling code, making their refactoring process feel unnatural compared to human developers.
- LLMs struggle with asking clarifying questions, often making assumptions and brute-forcing solutions instead of seeking clarification like human developers do.
- The behavior of LLMs in coding tasks resembles that of 'weird, overconfident interns' rather than replacing human developers, highlighting current limitations in their approach.