the brief · the standard · the pass · the check · the record
Nearly everybody learning to write with an assistant collects techniques. Techniques go out of date. What survives is an arrangement of five parts, and when the work comes back wrong it is almost always the same part that was missing.
A system is not a sequence. The days run in order because learning does; the parts do not, and any one of them can be the one you are missing.
Every one of the five is a decision about your own work rather than a way of talking to a tool, which is why none of them expires when the tools change.
People look for: AI writing workflow, writing standards, revision process, quality checks, and decision records.
Read down the last column. Whichever line describes what went wrong last time is the part you are missing, and the day beside it is where that part gets built.
| The part | What it decides | What it looks like when it is missing |
|---|---|---|
| The brief days 2, 12, 13 | Who it is for, what it claims, what it argues against | A draft that is fine and about nothing. You reorder it three times looking for the argument |
| The standard days 6, 11 | The rules, each carrying its own test | Every piece is good and no two are the same. Nobody can take over from you |
| The pass days 13, 14 | One job per instruction, markable afterwards | You ask for a better version and get a different one. Nothing accumulates |
| The check days 15, 23, 24 | One rule per look, plus what only you can verify | Something plausible reaches a reader and comes back as a question you cannot answer |
| The record days 20, 25 | What changed, what you checked, what you could not close | A settled decision reopened in six months, by somebody who was not there |
Almost nobody is missing all five. Most people have three, and the two they lack are nearly always the first and the last. The brief, because starting feels like progress and deciding does not, and the record, because by then the work is finished and the reward has already been collected.
The assistant performs a pass inside the system; the brief, standard, verification, and final decision remain human responsibilities.
A checklist is run once, at the end, by somebody hoping to find nothing. A system decides what gets made before it exists, and most of its work happens where there is still nothing to check. That is why the brief is the part people skip and the part that costs the most.
Each of the five points at somebody. The brief points at your reader, the standard at whoever takes over from you, the check at whoever signs it off, and the record at somebody who has not arrived yet. Nothing in the system points at the writer, which is what makes it a system rather than a routine.
Inside one part, and not the important one. The assistant does the pass. It cannot decide the claim, cannot know which rules you hold, cannot verify anything outside the text, and cannot say what it failed to close. Four of the five parts are yours whatever you are using.
Three writers used the same assistant and produced three incompatible voices. A shared standard with checks fixed the work across every assignment.
A polished draft was reorganised repeatedly because nobody had settled its reader or claim. Once the brief named both, the structure became obvious.
A disputed sentence had passed several reviews, but the source, change, and approver were absent. A four-line record would have preserved the reasoning.
It is a repeatable arrangement of five decisions: the brief defines the job, the standard defines acceptable work, the pass changes one thing, the check tests the result, and the record preserves what happened. The assistant mainly helps with the pass; responsibility for the other parts remains with the people doing the work.
A prompt is one instruction used at one moment. A writing system decides what the work must achieve, which rules survive across assignments, how revisions are separated, how claims are verified, and what is recorded. Prompts can change with the tool while those decisions remain stable.
Name the reader, the change the writing should produce, the central claim, the evidence available, the position it argues against, and any boundaries that cannot move. A brief should settle decisions the assistant cannot infer safely rather than describe every sentence in advance.
A rule such as 'be concise' leaves room for different interpretations. A rule such as 'no paragraph exceeds sixty words' can be checked. Tests turn preferences into standards, make reviews consistent, and reduce drift when a different person or model performs the next pass.
Record the source version, the model or tool used when relevant, the changes requested, the checks completed, unresolved questions, and the person who made the final decision. Keep it short and write it while the work is current; reconstructed records are less reliable.
A system is what is left when you stop remembering to be careful.