A “minimalist-centaur” playbook
Human = concept engine | AI = content engine
| Stage | Human’s 5-minute job | AI’s automated job |
| 1 · Spark | Write a one-sentence creative intent (“A sci-fi movie poster with Khmer temples in neon rain”). | — |
| 2 · Frame | Add guard-rails: purpose, tone, constraints (“Must feel hopeful, include 2025, print-ready A2”). | — |
| 3 · Prompt | Combine Spark + Frame into a structured prompt. | Parse prompt → generate first draft (image, copy, code, etc.). |
| 4 · Critique | Rapid thumbs-up/down; jot why (“temple silhouette too dark”). | — |
| 5 · Iterate | Rewrite the weakest line of the prompt (or feed examples). | Regenerate, honoring feedback. |
| 6 · Ship / store | Select best take, check for policy/bias issues, hit publish. | Export in final format, tag metadata. |
Rule of thumb:
Give the machine the what and why; let it handle the messy how.
Quick example: newsletter banner
- Spark – “Weekly Bitcoin digest with a playful 8-bit vibe.”
- Frame – Colors match #F7931A (BTC orange); 600×200 px; include the word “HODL.”
- Prompt – “Create a 600×200 px banner in 8-bit pixel-art style, dominant #F7931A, saying ‘HODL WEEKLY’ with a pixelated rocket and rising BTC chart.”
- AI draft – Returns four variants.
- Critique – Variant 3 too cluttered; ask: “fewer stars, thicker font.”
- Final – Approve variant 2, download PNG, embed in Substack.
Total human time ≈ 7 minutes.
Why it works
- Division of labour: people excel at taste & intent; models excel at scale & speed.
- Rapid feedback loop: each prompt tweak is a zero-cost A/B test.
- Cognitive unload: no wrestling with Photoshop layers or boilerplate code.
Common pitfalls & fixes
| Pitfall | Fix |
| Vague prompt → dull output | Specify purpose, tone, style refs, and non-examples (“avoid stock-photo vibe”). |
| Iteration fatigue | Limit yourself to three prompt cycles; if it’s still off, rethink the idea. |
| Over-delegation | Keep a human QA checklist (brand voice, factual accuracy, policy compliance). |
| Homogenous results | Feed the model diverse style cues or seed images; fine-tune if it still cycles sameness. |
TL;DR
- Human sparks the idea in plain language.
- AI manufactures variations at lightning speed.
- Human curates, tweaks, and green-lights the winner.
Keep prompts crisp, feedback granular, and iterations short—and you’ll turn raw ideas into polished assets before the coffee cools.