THE
CLAUDE
PLAYBOOK
Ship real products with AI — the rules, rituals and small machines that turn a brilliant amnesiac collaborator into a compounding one.
In the making — EN & DE. One email when both editions ship. Nothing else, ever.
CLAUDE
PLAYBOOK
Admission is simple: you ship.
Measured, not marketed
The system this book teaches, running on itself — our own logged sessions:
of input tokens served from prompt cache
lower input cost than the same work at uncached prices
more work for the same money — the cost ratio vs uncached
of all tokens are output — lean answers, by discipline
Measured — regenerated from the session logs on every build, never typed by hand.
Honest footnote: this measures cache efficiency — the same logged work priced with and without caching. It is not an A/B against working without this system; that number was never measured, so it isn't claimed.
One volume, two editions
Both ship on the same day. Pick the runtime you actually run — or take both and let them verdict each other.
The Claude Playbook
Maps your codebase can explain, sessions that boot, proofs instead of promises, ship discipline — and the runtime that runs the day and writes the memory.
The Grok Playbook
The same spine on the other runtime: what breaks differently, what a refused call does to a whole run, and when to bring Claude in as the verdict.
Same spine — chapters 1, 6, 7 and 10 are identical on purpose, and both books say so. What differs is the half that is really runtime-specific: Claude runs the day and writes the memory · Grok brings the speed and the contradiction. Two identical blurbs would be the scam.