AI 编码智能体效率低下，皆因工具输出过于冗余

AI coding agents exhibit significant performance bottlenecks due to verbose tool outputs, rather than inherent intelligence deficiencies. For instance, an AI code validator agent took 608 seconds to report results from a test suite that completed in just 96 seconds. This disparity arises because traditional developer tools—including test runners, linters, compilers, and build systems—are primarily designed for human interpretation of terminal output. When an AI agent processes this same output through its context window, unexpected inefficiencies and breakdowns occur. The article illustrates this problem with a TypeScript monorepo containing approximately 12,000 tests across four packages. Despite the agent's simple task of running tests and reporting pass/fail status with coverage, the time taken by the agent far exceeded the actual test execution time. This highlights a critical need to re-evaluate and optimize tool output formats specifically for AI consumption, ensuring conciseness and structured data to enhance agent efficiency and accuracy in code validation and other development tasks.