Our CTO argues that context windows are now a million t...

The argument that RAG is a transitional hack is flawed because it ignores the specific bottlenecks of AI systems. While long context windows allow for easy updates and better attribution, they convert a one-time index cost into a recurring, expensive per-query prefill cost that scales with context length. Furthermore, long context does not solve the opt-out or data deletion requirements. Fine-tuning remains a limited tool that only addresses the update bottleneck while leaving attribution, deletion, and privacy concerns unresolved.