RTK hook makes coding agents more expensive, not cheaper (blog.jetbrains.com)

A recent evaluation of the "Rust Token Killer" (rtk) hook for coding agents revealed that instead of achieving the promised 60-90% reduction in token usage, it actually resulted in a 7.6% increase in costs when integrated with the AI model Claude Code. RTK is designed to intercept and compress shell command outputs for more efficient processing. However, the study highlighted that much of the agent's output comes from built-in tools that bypass the rtk hook altogether. Consequently, the effective savings in token consumption were negligible, as the majority of command outputs were either not captured by rtk or already optimized by Claude Code's internal mechanisms. This finding is significant for the AI/ML community as it calls attention to the need for transparency and rigorous evaluation of claims made by AI tools regarding efficiency and cost-effectiveness. The study emphasizes that self-reported savings from tools like rtk can be misleading, as they may not reflect actual impact on performance or costs. This highlights the importance of conducting thorough benchmarking and analysis before integrating such tools into workflows, allowing developers to make informed decisions based on real data rather than promotional metrics.