The problem
Writing consistently about AI in HR meant the bottleneck was never ideas, it was turning a rough thought into something publishable without losing the voice and argument quality across every piece. A generic “ask an AI to write a blog post” workflow produces generic output.
What I built
A content pipeline in Claude Code (the FTPD-Content-Repo) built around two custom skills:
- article-writer — takes an idea or a set of notes and produces a full Substack article plus a paired LinkedIn teaser, following a fixed voice guide (no em dashes, no colons in prose, no stacked rhetorical questions, plain-spoken and specific) and a consistent format (Thought Doership and Thought Leadership pieces with dynamic subheadings).
- argument-research — pressure-tests a thesis before it goes in an article, running live searches to surface real counterarguments and steelman opposing views rather than letting a draft go out unchallenged.
TODO: This is the page most worth expanding — walk through what a single idea-to-published cycle actually looks like end to end, and consider a screenshot of the repo or the skill running.
What happened
The signature ideas that came out of this system, HR sequencing AI adoption backwards, AI as a reasoning partner rather than an oracle, and the cognitive load of human-in-the-loop AI, are now the core positioning anchors for FTPD. Growth has been driven primarily through LinkedIn.
TODO: Add current subscriber count and any other metric worth showing (open rate, LinkedIn engagement, which piece performed best).
What I’d do differently
TODO: A real trade-off — e.g., the pipeline enforces consistency but can also enforce sameness if you're not careful, or it's faster than writing from scratch but review time didn't shrink as much as expected.