This system turns one raw idea into research, strategy and channel-specific drafts while keeping a human approval gate before anything is published.

Workflow map

Idea intake
→ Research Agent
→ Strategy Agent
→ Writing Agent
→ Channel Agents
→ Creative Agent
→ Review Agent
→ Human approval
→ Publishing queue

Agent responsibilities

Agent Input Output
Research Raw idea and audience Claims, examples, sources and objections
Strategy Research packet Angle, thesis and content brief
Writing Approved brief Core argument and proof structure
LinkedIn Core argument LinkedIn-native post
X Core argument Thread or concise post series
Newsletter Core argument Long-form issue
Carousel Core argument Slide-by-slide outline
Video Core argument Short-form script and shot plan
Creative Channel drafts Visual direction and asset list
Review All outputs QA report and approval package

n8n data structure

Use one shared item throughout the workflow:

{
  "idea_id": "",
  "raw_idea": "",
  "audience": "",
  "objective": "",
  "research": [],
  "angle": "",
  "thesis": "",
  "proof": [],
  "channel_outputs": {},
  "visual_brief": {},
  "review": {},
  "status": "draft"
}

Node sequence

  1. Form or webhook trigger receives the idea, audience and objective.
  2. Set node creates the canonical data object and ID.
  3. Research model node returns structured research with source URLs and uncertainty flags.
  4. Strategy model node selects one angle and explains why it fits the audience.
  5. Writing model node creates the core argument.
  6. Parallel branches generate LinkedIn, X, newsletter, carousel and video outputs.
  7. Merge node combines channel drafts.
  8. Creative node creates the visual brief.
  9. Review node checks claims, repetition, tone and channel fit.
  10. Human approval step sends the package to Notion, Slack or email.
  11. Approved branch moves assets to the publishing queue; rejected items return with notes.