Solving the Wrong Problem
a day ago
- #AI-coding
- #LLMs
- #software-development
- AI-assisted and AI-based coding are discussed, with AI-assisted involving human coders with AI support and AI-based involving AI agents handling coding tasks.
- AI agents can create solutions but often require corrections and reminders, showing both potential and current unreliabilities.
- There's a concern that AI-based coding might not address the right problems, focusing on what's possible rather than what's needed.
- LLMs function based on probabilities and training data, leading to questions about creativity and the reuse of existing code fragments.
- The prevalence of crappy code is attributed to a lack of education and training among software developers.
- AI agents are seen as a way to speed up development but may not improve code quality, as they replicate patterns from their training data.
- The demand for faster feature delivery is criticized as potentially leading to more waste rather than value.
- Better documentation and requirements are ironically seen as necessary for AI agents, though they were often neglected for human developers.
- The article concludes that AI-agent-based development might be solving the wrong problem, cementing current issues rather than addressing them.