A working agreement with yourself

Decide before you are tired

Every hard call about using these tools arrives at the worst possible moment: eleven at night, deadline tomorrow, three hundred words short. Nobody makes good decisions there. So the decision has to be made now, in writing, while nothing is at stake.

This lesson asks you to write four short things. It takes fifteen minutes and it is the most useful fifteen minutes in this module.

One: the list of things you will not outsource

Name the specific capabilities your degree or your career is supposed to leave you with. Not subjects. Capabilities.

A medical student might write: taking a history, reading an ECG, calculating a paediatric dose. A computer science student: writing a function from a spec without help, reading a stack trace, reasoning about complexity. A history student: reading a primary source and forming an argument about it, structuring a paragraph. A language student: producing a sentence in the target language without a prompt.

These are the things you do unaided, always, even when it is slower. The list should be short (five or six items) because a long list is one you will abandon. Everything not on the list is fair game.

The test for whether something belongs on the list: if you cannot do it, can you tell when somebody else has done it badly? Supervision is most of what is left of many jobs, and you cannot supervise a skill you never acquired.

Two: your disclosure default

Decide now what you will write in a disclosure statement, and decide it on the generous side. Something like:

AI use: I used a language model to generate practice questions, to critique my draft structure, and to check my understanding of X. All prose, all argument and all sources are my own; I opened every source cited.

Writing the sentence in advance changes behaviour, because a use you would be embarrassed to disclose is a use you now notice yourself making. Module 7 covers institutional rules in detail. This is the personal version, which binds you before the rules do.

Three: your check for specifics

One line: every name, number, date, quotation and reference gets verified against a source I opened. Written down, it is a rule. Unwritten, it is an intention, and intentions lose to deadlines.

Four: your evidence habit

Decide how you will be able to show your work is yours, before anyone asks. The cheapest version: write in a tool that keeps version history (Google Docs, Word with autosave, or plain files in git) and never paste a finished draft into a blank document. That is the whole habit. Module 7 explains why it matters more than any argument you could make after an accusation.

Why an agreement beats a rule

Institutional rules tell you what is punishable. They are necessarily crude, they differ between departments, and they are silent on most of what you do. An agreement with yourself covers the enormous space the rules do not reach: the moment you nearly asked for the whole paragraph, the citation you nearly did not open, the problem set you nearly copied because it was late.

Nobody will check this document. That is the point. The students who come out of the next few years able to do things are the ones who decided, in advance and for their own reasons, what they were going to keep doing themselves.

A worked example, because abstractions are easy to nod at

Here is a real-shaped agreement, written by a second-year engineering student:

Never outsourced: free-body diagrams, unit and dimension checks, writing a function from a spec, reading a datasheet, the argument of any report.

Disclosure default: I will state which sections had AI critique and which had AI drafting, and there will be no sections with AI drafting.

Specifics rule: every constant, standard number and reference gets opened. If I cannot open it, it comes out of the report.

Evidence: everything in git, committed at the end of every session, even when the session was bad.

Notice what it does not say. It does not ban the tool. It does not promise heroic restraint. It names four small mechanical commitments that can be checked by looking, which is the only kind of commitment that survives a bad week.

Revisit it

Put a date on it and read it again at the end of each term. The list should change. Things you deliberately learned unaided move off it, because you have them; new things move on. A list that has not changed in a year is a list you stopped reading.