토큰 압축의 착시: RTK에 회의적인 이유

RTK's 60-90% token compression metric measures only terminal output reduction, not actual API cost savings, while overlooking larger factors like context depth and internal reasoning tokens. More critically, silent failures in output parsing can cause AI agents to make poor decisions without complete diagnostic information like stack traces or compiler context. Without published accuracy benchmarks proving agent success rates are maintained, adopting RTK in production workflows carries operational risks that outweigh the promised cost savings.