“A workshop you never reorganise slowly fills with tools you no longer use, until you can't find the one you need.”
The defaults are not tuned for you
The source article counts 125+ keys in Claude Code's settings.json and notes the official docs cover roughly 40. Whether or not those exact numbers hold, the point stands: there are far more knobs than any onboarding flow surfaces. The ones that matter most tend to be buried in menus, mentioned only in GitHub issues, or left blank by default.
Two failure modes follow from leaving everything on default. Drift: memory and styles accumulate one-off corrections and stale context until Claude is optimising for the wrong version of you. Cost: thinking, caching, model choice, and connected tools quietly inflate token spend on work that didn't need them.
How this course is laid out
- Chapter 2, Claude.ai (8 settings): memory, thinking, styles, projects, search, web citations, connectors, incognito.
- Chapter 3, Claude Code (7 settings): the settings.json keys that move context budget and cost.
- Chapter 4, API & Console (3 settings): the highest-leverage cost controls, from prompt caching to rate limits.
- Chapter 5, Checklist & audit: the 18-point walkthrough plus a script to re-check it weekly.
For each setting you get the same four things: where it lives, what it does, why it matters, and the one-line fix. Read top to bottom, or jump to the surface you actually use.
In one line each
- Default Claude is tuned for onboarding, not for you: most high-impact settings are hidden or left blank.
- Two costs of ignoring them: output drift (memory/styles) and token waste (thinking, caching, model, tools).
- The course covers 18 settings across Claude.ai, Claude Code, and the API, each as where / what / why / fix.
Where to go next