Five-Layer AI Content Team
Coordinate specialized AI skills across the full content lifecycle
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 94%
The Five-Layer AI Content Team organizes content work into interoperable skills rather than one oversized prompt. An orchestrator accepts the user's request and selects the appropriate audience, voice, research, creation, enrichment, and review capabilities. Persistent audience profiles and platform-specific style cards provide context that every downstream skill can load. Research skills create ideas and talking points; production skills turn them into channel-specific drafts; enrichment skills add evidence, stories, and examples. Finally, published outputs and their performance metrics flow into a recurring review that updates the underlying skills. The mechanism turns content creation into a modular system whose components can be replaced, extended, or automated without redesigning the entire operation.
Origin
Extracted from Marketing Against The Grain, where the host demonstrates an 11-skill Claude Code content team built across five layers.
Core principles
- 01Treat content production as a connected system, not isolated prompts
- 02Give specialized skills distinct responsibilities
- 03Centralize interaction through an orchestrator
- 04Ground every output in audience and voice context
- 05Close the loop with real performance data
How to run it
- 1
Map the lifecycle
Break the content process into repeatable responsibilities such as audience research, ideation, drafting, enrichment, and performance review.
Pro tip Use separate skills when responsibilities require different inputs or evaluation criteria.
Watch out Do not reproduce a manual workflow as one giant prompt.
- 2
Build the context layer
Create persistent audience profiles and writing-style files that downstream skills can load before generating content.
Pro tip Maintain separate style cards when your voice or format changes materially by platform.
Watch out Generic persona documents may describe buyers without capturing the content they actually engage with.
- 3
Add research skills
Create skills that discover trends, extract talking points, and identify ideas resembling proven content patterns.
Pro tip Allow both live web research and user-supplied source material.
Watch out Research without audience context can produce interesting but irrelevant ideas.
- 4
Add production and enrichment skills
Turn selected ideas into platform-specific drafts, then add evidence, stories, examples, or stronger hooks through dedicated enrichment skills.
Pro tip Keep drafting and enrichment separate so each can be evaluated independently.
Watch out Do not treat generated first drafts as publish-ready work.
- 5
Install an orchestrator
Provide one interface that inspects available assets, asks for missing inputs, and invokes the correct specialized skills in sequence.
Pro tip Let users refine decisions conversationally instead of forcing rigid workflow forms.
Watch out An orchestrator cannot compensate for unclear contracts between skills.
- 6
Capture performance
Store generated content alongside platform metrics so the system can connect outputs to outcomes.
Pro tip Automate metric collection where an API is available, but support manual entry as a fallback.
Watch out Do not optimize using unnormalized or incomplete metrics.
- 7
Run the improvement loop
Review performance on a fixed cadence and update the relevant profiles, patterns, and skill instructions based on what worked.
Pro tip Keep a record of why each skill changed so weak updates can be reversed.
Watch out Avoid rewriting skills from tiny samples or one unusually successful post.
In the wild
A creator stores an audience profile and separate Substack, LinkedIn, and X style cards. The orchestrator selects a researched talking point, drafts it for the requested channel, invokes an enrichment skill for evidence, and records the published result. At month-end, performance data updates the hooks and formats favored by each drafting skill.
→ The creator gains a reusable operation that produces contextual drafts and improves from measured results.
Common mistakes
Stopping at content generation
Drafting skills create output but do not create a learning system. Without performance capture and review, the same weaknesses persist.
Using one context for every platform
A single voice profile can flatten platform-specific differences in format, audience expectations, and successful patterns.
Automating before defining handoffs
Autonomy magnifies ambiguity when skills do not have clear inputs, outputs, and responsibilities.
Is it for you?
Best for
It is best for creators and marketing teams producing recurring content across several platforms.
Not ideal for
It is not ideal for occasional users who only need isolated drafts and have no performance data or repeatable workflow.
From the transcript
“It's actually 11 skills across five different layers of content.”
“So, the orchestrator skill is going to actually use all the other skills, which makes it really easy to use the system.”
“And then feedback loops. This is so important. If you're a system thinker, most people stop here.”
From the episode
My 11-Skill AI Content Team (Built in Claude Code)