Articles on how frontier AI systems work and how to use them well. Most have interactive figures; they read best on a laptop, and work on a phone.
The Cache Is the Bill
How prompt caching works inside frontier LLMs, and what it does to the cost of long agent sessions: cold misses, keep-alive pings, 5-minute versus 1-hour caches, forks, and a simulator for a whole working day on Claude Opus 5.5.
28 September 2026 · LLM inference, agents, cost · 8 interactive figures