I thought the next halving was due for 2028… Tenki moves different 👀
The last few days we've been grinding to halve our ai code reviewer time to completion.
- P95: 15m -> 7m
- P50: 10m -> 4:30m
We did so w/o sacrificing intelligence or accuracy.
The speed gains come from working with our inference partners to dial in tokens per second throughput, optimize cache hits in the harness and heavily optimizing the workflow.