📄 中文摘要
大多数 AI 代理假设人类是理性的，但实际上只有 24% 的决策依赖于逻辑，其余则由情感和直觉驱动。许多 AI 项目在构建智能系统时存在五个常见的误区：首先，过于简化人类行为，假设决策是线性的，而人类的决策受到复杂情感的影响；其次，忽视领域知识，依赖算法而非专家见解，可能导致严重错误；第三，低估上下文复杂性，未能考虑幽默、讽刺或文化差异等细微差别；第四，缺乏对人类情感的理解，导致 AI 无法有效互动；最后，未能进行充分的测试和迭代，导致系统在真实环境中的表现不佳。通过认识并解决这些问题，可以提升 AI 系统的智能水平和用户体验。
📄 English Summary
The 5 Things Most AI Agent Projects Get Wrong (And How We Fixed Them)
Most AI agents assume that humans are rational, but only 24% of our decisions are based on logic, with the remainder driven by emotions and intuition. There are five common pitfalls in many AI agent projects: first, oversimplifying human behavior by assuming linear decision-making, while human decisions are influenced by complex emotions; second, ignoring domain knowledge and relying on algorithms instead of expert insights, which can lead to catastrophic mistakes; third, underestimating contextual complexity by failing to account for nuances like humor, sarcasm, or cultural differences; fourth, lacking an understanding of human emotions, which hampers effective interaction; and finally, failing to conduct adequate testing and iteration, resulting in poor performance in real-world environments. By recognizing and addressing these issues, the intelligence and user experience of AI systems can be significantly improved.