20+ year senior software engineer here, sharing my experience with LLM/AI agents.
TL;DR: LLMs are genuinely useful and many experienced developers use them daily, but they’re being oversold. They work best as tools that amplify expert developers, not as replacements for them. They can generate and refactor code well, but they hit a hard ceiling on deep reasoning, architecture, and optimization. In expert hands, they save time; treated as autonomous engineers, they produce confident-looking but low-quality results.
TL;DR: LLMs are genuinely useful and many experienced developers use them daily, but they’re being oversold. They work best as tools that amplify expert developers, not as replacements for them. They can generate and refactor code well, but they hit a hard ceiling on deep reasoning, architecture, and optimization. In expert hands, they save time; treated as autonomous engineers, they produce confident-looking but low-quality results.