Your Bot does the job in four minutes. Checking its work takes you fourteen. The week has saved six minutes, which is not what you set it up for. That is the second stall, and no first-week advice reaches it. A Bot that works and a Bot that saves you time are two different claims, and what sits between them is your reading. This book is about earning the right to read less — deliberately, on evidence, in a way you could defend to somebody who was not in the room. It assumes you have had a Bot for a few weeks and it does its job. It stays with a single Bot from the first page to the last: group chats and handoffs are a different problem, for a different book. Nine chapters, each ending with work you can do the same afternoon. A skill is not a saved prompt. It is when to use it, what access it needs, the sequence of the work, how the result is checked, what comes back, and what has to be approved. Leave one of those out and it is the skill that breaks the first time a source returns empty. You will write one properly, then put it on a schedule — after you have tested it, because a routine can navigate sites, change files and call every tool you connected, on a Monday morning when you are not watching. Then the failures that raise no error. A session that quietly expired and left a cached table behind. A connector serving last week's page. A routine that ran against an empty input, reported success, and went on doing that for three weeks. You keep twenty run records per routine; that is the whole of your evidence, and it is not much, so the book shows you what to put in the deliverable instead. The review budget is the heart of it. Four acceptance criteria written into the request itself — a figure that must reconcile, a count that must match, a link that must resolve, a row that must be present — so a result that took twelve minutes to read is settled in one. A fortnight of full checks to earn the step down, including a blind recheck against the raw sources, because one agreement with the world is worth more than a fortnight of agreements with yourself. Then one run in five, sampled by dice rather than by convenience, with the triggers that send you back up named in advance. And a line none of it crosses: work that is irreversible, work that moves money, work that touches a person's record is checked every time, forever. Costs get the same treatment. Weekly usage is included, the add-on is billed on model and token cost, xAI publishes no figures, and neither does this book; you are shown how to tell an expensive habit from a useful one instead. Grok Bot went to beta in August 2026 and its plan list changed fifteen days later. Where this book and your screen disagree, trust the screen. Book two of The Grok Bot Guides, after Grok Bot for Beginners. One Bot, one month, and a page you can hand to whoever asks why you stopped reading every line.
