After months of coding with LLMs, I'm going back to using my brain (albertofortin.com)
Source: ycombinator.com
This Hacker News discussion revolves around the effectiveness of using LLMs for coding. Key points include:
- LLMs are useful for boilerplate, small functions, and learning new syntax.
- They can boost productivity, especially for junior developers or greenfield projects, but require careful review.
- Blindly relying on LLMs (\"vibe coding\") often leads to poor code quality and maintainability issues.
- Experienced developers find value when using LLMs judiciously as a tool, not a replacement for deep thought.
- Concerns exist about skill atrophy and the reliability of LLMs for complex tasks or less common technologies.
The main points include:
- LLMs as Tools, Not Replacements – The author acknowledges that LLMs can accelerate development but argues that they should be used strategically rather than relied upon completely.
- Balancing Automation and Understanding – Some developers use LLMs to generate non-critical components like landing pages or simple UI elements, while maintaining full control over the core application.
- The "10x Engineer" Debate – Different perspectives emerge regarding whether LLMs provide a major productivity boost. Less experienced developers may see a huge improvement, while seasoned engineers tend to integrate LLMs into existing workflows without dramatically altering their approach.
- Concerns About Code Quality – The discussion highlights that blindly accepting AI-generated code can introduce subtle bugs, which may be harder to detect later. Experienced developers emphasize reviewing and refining AI-assisted output.
- Learning vs. Over-Reliance – Some argue that junior developers might stagnate if they depend too much on LLMs for coding rather than learning deeper problem-solving and software architecture skills.
- LLMs Are Part of a Broader Evolution – The discussion compares current AI-driven coding tools to earlier innovations in automation and integrated development environments (IDEs), suggesting that AI will continue evolving rather than replacing traditional programming.
Overall, the conversation reflects a nuanced view: LLMs are valuable tools, but careful integration and maintaining fundamental coding skills are crucial for long-term success.