Creative Practice Delegation Map
A worksheet for deciding, stage by stage, what to accelerate with AI and what to keep deliberately human. Companion to Keep your taste: AI in a creative practice.
Map your actual process
Use one recent, real piece of work — not a generic template.
| Stage | What I currently do | What AI currently does (if anything) | Builds my skill? (Y/N) | Carries my voice? (Y/N) | Process is the product? (Y/N) | Decision: accelerate / protect |
|---|---|---|---|---|---|---|
| Idea generation | ||||||
| Reference/research | ||||||
| Drafting/sketching | ||||||
| Technical execution | ||||||
| Critique/editing | ||||||
| Finishing/polish | ||||||
| Distribution |
Rule of thumb: any stage with at least one “Y” is worth protecting even if AI could do it faster. A stage with all “N” is a reasonable candidate to accelerate.
Before/after critique prompt
Compare these two pieces of my own work: [describe or paste both].
One used AI assistance at [stage]; one did not.
Looking only at craft quality and distinctiveness — not effort or
speed — which one is more clearly and specifically "mine"? Which one
would a person who knows my work recognize faster as something I
made? Be specific about what creates or removes that recognizability.
Do not tell me which process was better overall. I am asking about
one narrow comparison.
Copyright and provenance record
- For pieces where ownership might matter later (licensing, sale, dispute), I kept a record of my own creative contribution: drafts, layers, revision history, or notes on what I changed.
- I know that, under current US Copyright Office guidance, purely AI-generated output is generally not copyrightable on its own — only my own creative modifications are.
- If I work under a different jurisdiction, I have checked whether its rule differs before assuming the US position applies.
Consent and imitation checklist
- I have not asked a tool to imitate a specific, identifiable living artist’s style for commercial use without considering a license, permission, or open attribution instead.
- I treat any claim about a model’s exact training data as unverified unless the provider has published a specific, checkable disclosure.
Disclosure checklist
- I have disclosed AI involvement wherever its absence would materially mislead my audience about what they are looking at or who made it.
- For deepfakes or AI-generated text on matters of public interest, I am aware of the EU AI Act’s Article 50 disclosure requirements (applicable from 2 August 2026) if they apply to my work or audience.
Periodic unaided check
| Date | What I made without any AI assistance | Did it feel harder/weaker than it used to? | What that tells me |
|---|---|---|---|
Run this every few months. A noticeably harder or weaker unaided attempt is a signal to move a stage from “accelerate” back to “protect” — not proof that AI use in general is a problem.