An AI content system,and the skill library behind it
01 — Summary
A full content lifecycle plus the AI skill library that runs it: drafting, automated review, human approval, publishing and logging, with one source of truth.
My role — System design, implementation and development of the AI skill library.
current iteration of the voice skill
automated engagement reporting
02 — The problem
Content was being produced the way it is produced almost everywhere: ad hoc, in whichever tool was closest, with quality depending on who happened to pick it up. Nothing was measurable, nothing was repeatable, and institutional voice lived in people's heads.
03 — What I did
- Mapped the real lifecycle from first draft to published and logged, including the approval steps people were doing informally.
- Built AI-assisted drafting into the front of the process and automated review behind it, keeping human approval and editing in the middle where judgment matters.
- Connected the tools by API so nothing depended on copy-paste, with a single shared source of truth for what had been published.
- Authored a library of AI agent skills to encode voice and process: a company-page voice skill now on its sixth iteration, separate skills for individual executive voices, a searchable quote bank, a newsletter assembler, and a campaign-tagging tool.
- Built the publishing pipeline itself: an automation that posts to LinkedIn and X through Typefully and announces each post in the right Slack channels automatically.
- Added skills that draft posts automatically, a shared content calendar, and WhatsApp automation for the parts of the process that happen there.
- Built executive thought-leadership workflows on top: WhatsApp automations, voice-based skills and structured processes that help experts share authentic perspectives consistently, rather than simply publishing more posts.
- Built an internal content map showing what is being created, what performs, and where the next opportunities are.
- Added automated daily engagement reporting and weekly competitive intelligence, both still running.
04 — What changed
- An ad-hoc process turned into a repeatable, measurable workflow with one source of truth.
- Voice consistency that does not depend on one person being available, and does not flatten into generic copy.
- Teammates able to produce work in the system without needing deep technical skill, which was the point of the whole exercise.
What I took from it
The interesting part was not v1, it was v6. A voice system that survives six iterations is one that has been corrected by real use rather than designed in the abstract. Anyone can build the first version.