Tutoring the Enemy No. 0 — Learning With an AI, in Both Directions

The method paper of Tutoring the Enemy, a series that starts from one question and works, with an AI, toward a shared understanding. It sets out the method followed on 1 November 2026, in a two-hour session with Claude (Anthropic) that produced Tutoring the Enemy No. 1, so that anyone can try it with the AI of their choice, on their own question. The method has seven steps: ask one broad question; push back, and accept the corrections the evidence brings in turn; name the level of understanding sought; weigh each example by the attention it gets, the people it touched and the quality of its evidence; build key definitions in the discussion and state them as such; keep a log of every correction, in both directions, and of every change in understanding, never overwritten; write up, then check for drift. The write-up opens with the questions and what we have learned so far, and keeps the process in an addendum. A drift watch addresses a typical failure of AI writing: when the machine compresses an argument into a memorable line, descriptions slide into values and metaphors into definitions. Two examples from No. 1 show it. The paper ends with a text to paste at the start of a session and five points to report after trying it. This is version 0.1 of the method, used once, by one person with one AI. It is offered for testing by other people, with other machines and other questions. Term proposed: the drift watch (worded by Claude at the author's initiative). Written in part with Claude (Anthropic), an AI model. The ideas, editorial decisions and responsibility for the content are the author's.