📄 中文摘要
在使用 GPT-4 构建自主代理两年后，作者转向 Claude Code，尽管两者在成本和速度上没有显著差异。切换的原因在于 GPT-4 在文件处理方面频繁出现微妙的错误，例如文件操作的幻觉、偶尔的写入截断，以及虚构的 API 调用。这些问题导致了代理的功能不稳定，最终促使作者寻找更可靠的替代方案。Claude Code 的使用体验让作者感到满意，尽管它并不具备更高的性能或更低的成本。
我从 GPT-4 切换到 Claude Code 用于自主代理。这实际上改变了什么。
📄 English Summary
I switched from GPT-4 to Claude Code for autonomous agents. Here's what actually changed.
After two years of building autonomous agents with GPT-4, the author switched to Claude Code three months ago, not due to cost or speed advantages, as both models are comparable in those aspects. The switch was prompted by persistent subtle issues with GPT-4, such as hallucinated file operations, occasional truncated writes, and fictitious API calls that led to instability in the agents' functionality. These challenges drove the author to seek a more reliable alternative. The experience with Claude Code has been satisfactory, despite it not offering enhanced performance or lower costs.