Tapri Rush
Two versions of one chai-stall game, each written by a different AI model (Fable 5.2, Opus 5.5) from the same prompt. Brew, serve, pay the kiraya. Works on phones.
topic
How agents, tools, caching and evals actually work, with the numbers.
Fable 5.2 and Opus 5.5 each wrote a complete browser game from the same one-shot prompt. A test bot played both. What each one built, and what it missed.
10 Nov 2026 · 5 min read
Every turn of an agent re-sends the whole conversation. Here's the formula for what that costs, and how much prompt caching takes off the bill.
14 Oct 2026 · 7 min read
in the lab
Two versions of one chai-stall game, each written by a different AI model (Fable 5.2, Opus 5.5) from the same prompt. Brew, serve, pay the kiraya. Works on phones.
What an AI agent costs per conversation and per month, with and without prompt caching.
the newsletter
The week's video, written out properly, with sources. Nothing else. Unsubscribe in one click.
Watch on YouTube ↗