The careful chat model, configured to externalise your executive function.

Claude is the chat model with the steadiest tone and the most reliable refusals — it pushes back instead of confidently inventing, which matters when the thing you are reasoning about is your own health or your own overloaded week. Its Projects feature is the part that turns it from a clever toy into a system: a Project keeps standing instructions and documents scoped to one job, so every new chat inside it already knows the moves. You do not re-paste anything. That persistence is the whole reason this beats a screenshot of seven prompts you will lose by Thursday.

Claude is the front door, not the whole house. Sitting underneath it is your ledger — a notes file or spreadsheet where the day's open loops, finished items and time-maps get written down so the structure lives somewhere permanent. Sitting alongside is a capture habit: when a thought or a symptom or a task lands, it goes into the same thread the same way you would message a friend. Once a week you ask Claude to export everything as a clean markdown list and paste it into the ledger. The model does the executive function; the ledger keeps the memory; you spend your attention on the doing.

The reason most people abandon a personal health system is not laziness — it is executive load. Remembering to log, deciding what matters, sequencing the weekly review: these are the exact jobs an under-resourced brain drops first. The same prompts that unstick a work task unstick a health habit. The open-loops dump becomes your Sunday review. The time-map stops you over-promising your evenings. Focus and health turn out to be the same problem wearing two outfits, and the same externalising workspace fixes both.

There is no special model for this, only a ladder you climb as far as the task needs. Haiku is the cheap, fast rung — point it at an export or a messy paste and let it pull the numbers out before anything reaches your ledger. Sonnet is the default you will live in: it runs the weekly review, the open-loops dump and the time-map without breaking a sweat, and it is the right answer for almost every entry. Opus is the top rung you reach for rarely — when a question genuinely needs hard cross-referencing across months of notes and Sonnet starts to stall. Set Sonnet as the Project default, drop to Haiku for grunt extraction, and save Opus for the once-a-month hard read. Match the rung to the job and the whole system costs you almost nothing.