Appendix H: The First Month
The practice manual, started. Four weeks, nothing heroic, every step from a chapter you have read. Tool names are deliberately absent; run this on whatever models you use, and it survives the next release.
Week one: reps. Route the ordinary day through a model, every day: the inbox summary, the meeting notes, the task list, the first draft you would have written cold. Expect friction; you are pouring the floor, not saving time yet. One habit starts now and never stops: at the end of each working day, ask the model what it learned about you and your preferences, and save the answer as standing context for tomorrow. The only metric this week is that you did it every day. (Chapter 4.)
Week two: divergence and the first loop. Keep the daily reps. Add one divergence run: a real question from your week, put to three or four different models, with five minutes spent on where they disagree, because the disagreement is the lesson. Then run one task, just one, as a loop instead of a faucet: you write the three-sentence spine, one model drafts, a different model judges, you decide what ships. Write the role split down on one page. That page is the first artifact of your record. (Chapters 4 and 5.)
Week three: the interview and the battery. Two sittings. First, let a model interview you for an hour about how you actually work: what you delete on sight, what makes you stop trusting an expert, which trade-offs you always resolve the same way. Save the answers; that document is the seed of your configuration file, and writing it teaches you more than the model. Second, run your first battery: five or six ambiguous prompts, a contradiction, an out-of-scope request, a case that deserves “I am not sure,” against the model you use most, and note what it defaults to. That note opens your dossier. (Chapters 6 and 7.)
Week four: the first check. Pick one live judgment that is yours this week, an estimate, a priority call, a review. Write your answer in three sentences, timestamped, before you let the model touch it. Then compare, and log who was right when reality reports back. That is one flip card, and it is the first line of your Proficiency Record (Appendix C). Before the month ends, do two more things: ask one colleague to be your partner for the monthly planted-defect exchange, and write down your demotion thresholds for the coming quarter, while nothing is wrong and the numbers are easy to be honest about. (Chapter 8.)
What you hold on day thirty. A loop page, a dossier opened, a configuration seeded, a record begun, a partner named, thresholds dated. Six entries, all real, all yours. The month did not make you better with AI; it made you measurable, which is the precondition for better. The lists in your feed will keep arriving. You now have the apparatus that turns any of them into a rep, and a record that will still mean something when every tool they mention has been renamed.