📄 中文摘要
AI 编程工具在处理复杂任务时面临显著局限，核心问题在于上下文理解不足。Copilot 等工具擅长简单的代码补全，但难以理解涉及多重业务逻辑（如支付服务中基于区域、账户年龄和订阅等级选择提供商）的复杂场景。这种局限性并非 Copilot 独有，而是所有 AI 编码工具（如 Cursor、Cline、Claude Code）的普遍问题。AI 编程工具的主要瓶颈体现在跨文件修改能力不足。通常，一个功能需求会涉及 5 到 15 个文件，分布在 3 到 4 个目录中，而 AI 工具往往只能看到当前文件或少数相邻文件，无法获取全局视图。例如，一个需要修改 API 端点、数据库迁移和服务层的任务，AI 难以全面理解并协调这些跨文件、跨模块的改动。这种缺乏整体上下文的缺陷，导致 AI 无法有效处理需要深层业务理解和系统级协调的复杂编程任务。
📄 English Summary
Why Copilot Doesn't Work on Your Hardest Tickets
AI coding tools, including Copilot, struggle with complex programming tasks primarily due to a fundamental context problem. While proficient at simple code auto-completion, these tools fail to grasp intricate business logic, such as a payment service's multi-faceted provider selection strategies based on customer region, account age, and subscription tier. This limitation is not unique to Copilot but extends to all AI coding assistants like Cursor, Cline, and Claude Code. A major breakdown point for these tools is their inability to handle cross-file changes effectively. A typical feature ticket often requires modifications across 5 to 15 files spanning 3 to 4 directories. AI tools generally only perceive the current file, or a few adjacent ones, lacking the comprehensive overview necessary for such tasks. For instance, an experiment involving a ticket that demands changes to an API endpoint, a database migration, and a service layer highlights this deficiency. The AI cannot synthesize the full picture, understand the interdependencies, or coordinate the required modifications across disparate parts of the codebase. This lack of holistic context prevents AI from tackling complex tickets that demand deep business understanding and system-wide coordination.