RESEARCH
AUGUST 2026
The same token has more than one price
Batch tiers, priority tiers, off-peak windows, and provider spreads mean identical tokens routinely differ 4x in price. Most teams pay the top of that range without deciding to.
Notes from the
execution frontier.
Research and engineering writing on inference economics, execution policies, and proving changes with evidence.
ALL POSTS
Inference economics, evidence pipelines, and the pricing surface under production AI.
EVIDENCE
AUGUST 2026
Proof, then traffic
How offline replay against your own captured traffic proves a cheaper execution policy before a single production request moves.
STRATEGY
JULY 2026
Your model is one decision. Your execution policy is dozens.
Model choice is the most visible knob on an AI workload — and rarely the largest one. Everything decidable at the API boundary is one policy, and it compounds.