Build Your AI Agent for Content Creation
The three-file agent architecture behind our own content pipeline, and the automation rules that decide what's worth building.
Members: watch the recordingKey takeaways
- The 3-file agent architecture
- The 10x automation rule
- Design-to-code pipeline with HTML templates
- Autonomous execution vs. manual approvals
- Practical tips and hacks
The 3-file agent architecture
Structure your agent's context into three Markdown files. claude.md: master instructions governing what the agent can and can't do. business_context.md: your background, mission, target audience, core themes. content_strategy.md: brand voice keywords, structures, positive guidelines, negative constraints.
The 10x automation rule
If a manual task happens more than 10x a week (event listings, social banners, recurring updates), it's worth automating via agentic APIs and structured prompts.
Design-to-code pipeline with HTML templates
Convert design mockups, PowerPoint slides or Figma layouts into clean HTML templates so your agent can dynamically inject fresh data without breaking the visual layout.
Autonomous execution vs. manual approvals
For recurring scheduled agents, set approvals to 'skip all approvals'. Leaving it on manual approval causes background runs to pause or fail.
Practical tips and hacks
Feed raw voice notes and meeting transcripts in rather than pre-edited text. Unfiltered thoughts carry richer brand nuance. Toggle Claude Code Plan Mode before complex tasks to save tokens. Connect the Canva API, and use Gemini's native image generation for higher-fidelity graphic layouts. Route agent outputs and review reminders onto your calendar as scheduled blocks rather than a to-do app.
What members walked out with
The pipeline demoed in this session now produces our own event recaps.
Bonus power prompts
Design screenshot to HTML template
“Convert this screenshot into an HTML file without any mock data, and save it as a standing template that you dynamically populate for me with new inputs.”
Raw thought context extractor
“Analyze this raw transcript of my spoken thoughts and extract: 1) My core mission and goals, 2) Target audience persona, 3) Brand voice keywords, and 4) Negative constraints (what to avoid). Output these into structured business_context.md and content_strategy.md files.”
