Show HN: I measured resuming an AI coding session: 22,897 tokens vs. 1,013 (vericommand.net)

A recent measurement conducted by VeriCommand revealed that resuming AI coding sessions using a record can drastically reduce token consumption, demonstrating a reduction from 22,897 tokens to just 1,013 for a specific multi-file task. This benchmarking, which accurately tallies every token involved, highlights the efficiency of using a compact, chain-verified state block to resume work instead of re-reading all previously touched files. The study found that the small overhead of maintaining this record—around 427 tokens per session—becomes effective as soon as the session crosses a boundary or when multiple models are involved. This finding is significant for the AI/ML community as it illustrates not only substantial savings in computational resources but also the potential for more efficient handling of complex coding tasks that span multiple sessions. The data suggests that the record-based approach is particularly beneficial in scenarios where context switching is frequent, potentially leading to improvements in overall productivity and cost-effectiveness in token utilization. By offering a viable solution for managing long-running AI processes, this approach may encourage further exploration and adoption of more efficient state management strategies in AI development.