AI WRITING AUTOMATION FIRST-MOVE CARD Use this card to decide whether to automate drafting first, automate editing first, or stabilize the work before automating either stage. ============================================================ 1. WRITING JOB ============================================================ Writing job: ____________________________________________________________ Audience: ____________________________________________________________ Finished artifact: ____________________________________________________________ Current owner: ____________________________________________________________ Approval owner: ____________________________________________________________ What must remain human-led? ____________________________________________________________ ____________________________________________________________ ============================================================ 2. DRAFTING LANE ============================================================ Current human drafting effort: ____________________________________________________________ Quality of the source packet: [ ] Weak [ ] Mixed [ ] Strong Brief quality: [ ] Vague [ ] Partly defined [ ] Clear and repeatable Output structure: [ ] Highly variable [ ] Partly repeatable [ ] Highly repeatable Transformation distance: [ ] Low [ ] Medium [ ] High Ease of verifying the finished draft: [ ] Difficult [ ] Manageable [ ] Easy Human judgment that remains necessary: ____________________________________________________________ AI-created repair: ____________________________________________________________ Existing human work that remains unchanged: ____________________________________________________________ Worst observed failure: ____________________________________________________________ Was the failure easy to detect? [ ] No [ ] Partly [ ] Yes Recovery path: ____________________________________________________________ ============================================================ 3. EDITING LANE ============================================================ Quality of the human source draft: [ ] Weak [ ] Mixed [ ] Strong Editing work is: [ ] Highly variable [ ] Partly repeatable [ ] Highly repeatable Intended meaning is already present: [ ] No [ ] Partly [ ] Yes Transformation distance: [ ] Low [ ] Medium [ ] High Can before-and-after changes be inspected? [ ] No [ ] Partly [ ] Yes Human judgment that remains necessary: ____________________________________________________________ AI-created repair: ____________________________________________________________ Existing human work that remains unchanged: ____________________________________________________________ Worst observed failure: ____________________________________________________________ Was the failure easy to detect? [ ] No [ ] Partly [ ] Yes Recovery path: ____________________________________________________________ ============================================================ 4. FINISHING-COST LEDGER ============================================================ DRAFTING LANE Preparation: ________________________________________________ Generation: __________________________________________________ Verification: ________________________________________________ Repair: ______________________________________________________ Approval: ____________________________________________________ Maintenance: __________________________________________________ Total finishing cost: ________________________________________ EDITING LANE Preparation: ________________________________________________ Editing: _____________________________________________________ Verification: ________________________________________________ Repair: ______________________________________________________ Approval: ____________________________________________________ Maintenance: __________________________________________________ Total finishing cost: ________________________________________ Do not treat fast generation or editing as savings until the candidate output reaches the accepted completion standard. ============================================================ 5. FAILURE SHAPE ============================================================ DRAFTING LANE Does failure appear visibly? [ ] No [ ] Partly [ ] Yes Can the reviewer detect it quickly? [ ] No [ ] Partly [ ] Yes Can the work return to a clean state? [ ] No [ ] Partly [ ] Yes Could failure trigger downstream action? [ ] No [ ] Yes Largest plausible repair: ____________________________________________________________ EDITING LANE Does failure appear visibly? [ ] No [ ] Partly [ ] Yes Can the reviewer detect it quickly? [ ] No [ ] Partly [ ] Yes Can the work return to a clean state? [ ] No [ ] Partly [ ] Yes Could failure trigger downstream action? [ ] No [ ] Yes Largest plausible repair: ____________________________________________________________ ============================================================ 6. LOSING-LANE TEST ============================================================ If editing is expected to win, test a source draft containing a subtle factual weakness. Did editing preserve or strengthen the error? [ ] No [ ] Yes What happened? ____________________________________________________________ If drafting is expected to win, remove one important fact from the source packet. Did drafting expose the gap or invent around it? [ ] Exposed the gap [ ] Invented around it [ ] Other: __________________________________________________ Additional test: [ ] Conflicting sources [ ] Important qualification [ ] Changed date [ ] Assignment that should be rejected or escalated Result: ____________________________________________________________ ============================================================ 7. DECISION ============================================================ [ ] AUTOMATE DRAFTING FIRST The inputs and output shape are stable enough that generation burden is the larger opportunity. [ ] AUTOMATE EDITING FIRST Useful source drafts already exist, transformation is bounded, and changes are easier to inspect. [ ] AUTOMATE NEITHER YET The underlying stage is too unstable to automate responsibly. Reason: ____________________________________________________________ ____________________________________________________________ Evidence used: ____________________________________________________________ What would change this decision? ____________________________________________________________ Review date: ____________________________________________________________ Decision owner: ____________________________________________________________ RULE TO REMEMBER Automate the stage where mistakes are easiest to see, contain, and correct.