AI-Augmented Workflow

Ship With Intent is itself a demonstration of Intent Engineering. The production workflow uses AI at every step where it adds value — and stops where it doesn't. The human handles the thinking. The AI handles the repeatable parts.

The Interview Step

Every piece starts with a Claude-assisted interview. Not "write me a blog post about X." A structured conversation where Claude asks questions to find the real point — not the obvious one.

The interview generates structured output:

• Core Insight — One sentence. The thing the piece proves.
• Supporting Points — Five numbered points that build toward the insight.
• Hook Angles — Two or three alternative opening approaches.

This step exists because the first idea about what a piece is about is almost never the actual point. The interview process forces the insight to surface through conversation, not assumption. The writer knows what they care about but often hasn't articulated why it matters to someone else.

Outline and Approval

The interview notes become a bullet-point outline following the seven-step structure arc. The outline must be approved before any script or draft writing begins.

This gate matters. Writing without a locked outline produces draft after draft of exploratory prose that never converges on a point. The outline locks the structure. Everything after is execution.

Script and Draft Generation

For video: a full spoken-language script, double-spaced between beats. The test is "read it aloud." If it doesn't sound like a phone call to a respected peer, it gets rewritten. The AI generates the initial script. The human reads it, marks what sounds unnatural, and the AI revises.

For written content: the draft follows the structure arc directly. The AI writes the first version. The human checks it against the quality checklist — single clear point, grounded example, analogy placement, earned conclusion, no padding. Revision is collaborative: the human identifies the problem, the AI generates alternatives.

Metadata Generation

After content is locked, Claude generates platform-specific metadata:

For YouTube:

• Five title options (pick one, max 70 characters)
• Description blocks: visible teaser (150 chars), content bullets, chapter markers, links, boilerplate
• 12-15 tags, ordered from specific to broad
• Three-word thumbnail copy options

For Substack:

• Email subject line (distinct from the title — this is the hook)
• Tags
• Social sharing description

For LinkedIn:

• Adapted post (150-300 words)
• Engagement question for the closing

The metadata step is fully AI-driven. The human picks from generated options. This is exactly the kind of work where AI outperforms — generating variations of structured content with specific constraints.

The Review Gate

Before publication, every piece goes through a review command (/review-content) that checks it against four specification documents:

• Brand voice — Peer-to-peer register, no inspirational padding, no performed insight
• Legal guidelines — No employer name, proper attribution, FTC compliance
• Content standards — Structure arc complete, hook test passed, quality checklist clear
• Platform requirements — Format matches the target platform, metadata complete

The review is AI-executed against human-written specs. It catches naming violations, structural gaps, legal risks, and quality issues before a human sees the piece. This doesn't replace editorial judgment — it handles the mechanical checks so editorial judgment can focus on the parts that require taste.

What the AI Doesn't Do

The AI doesn't choose the topic. It doesn't decide the point. It doesn't determine what matters. It doesn't know if an analogy is good — only if one is present. It doesn't know if the voice sounds right — only if the checklist criteria are met.

The creative work — choosing what to say and why it matters — stays human. The production work — structuring, formatting, metadata, compliance checking — is AI-augmented. Running the operation this way is itself a demonstration of the thesis: AI accelerates the repeatable parts so the human parts can stay human.