-
5 Sessions, built from one discovery call21 Needs mapped out of that call0 Sessions spent teaching prompting70/30 AI does 70, the human finishes the 30
-
The problem
The creative and brand team at a US consumer brand were already AI first movers. They did not need a beginner course, and they would have hated one. What they had was a harder problem: every output started from zero. Someone had to re-explain the brand, the tone, the things they are legally not allowed to claim, every single time, to every tool. Their influencer picking was done by eye, a profile at a time, hoping nobody had a scandal in their past. And they had no idea what their competitors were actually doing beyond scrolling the feeds.
-
The solution
Five sessions, built entirely around their real work. The flagship one turned their brand voice into a reusable Skill, so every output starts on brand instead of being dragged there. Then ideation, content pipelines that turn one piece into many, and ops automation for the email and project management sludge. Out of those sessions came working engines they still run: an influencer vetting engine that scores a creator out of 100 against their own rubric and runs a background check before anyone signs anything, a competitor intel engine that scrapes and labels rival content and hands back the cadence, formats and engagement they are up against, and a video generation setup that puts a hundred variations within reach of an afternoon instead of a six week shoot.
-
How it was built
The brand voice became a Skill: their tone, their per-platform gradations, their do-not-say list, encoded once and shared across the team. The influencer engine is a scored pipeline: creators pulled from their discovery platform, filtered on hard rules, unverified numbers re-checked with scrapers, then scored on a weighted rubric where audience fit carries the most weight, then a web grounded background check, then a dossier. The competitor engine scrapes both platforms across three brands, labels every post with an AI pass, and aggregates cadence and engagement into a report. The video work runs through an MCP connector, so Claude can drive the generation engines by conversation instead of the team tab hopping between five sites. The rule under all of it: the engine does the boring 70 percent, a human makes the final call. It never signs anyone and it never sends anything by itself.
Watch the wires draw. This is the real architecture.
Claude SkillsClaude CodeMCP connectorsApify scrapersPython and Excel
WANT ONE OF THESE FOR YOUR BUSINESS?
Schedule a Call
Contact