Parses your page and reviews; separates verified claims from inference; flags what needs a source.
Builds personas from real language, then kills every angle candidate that is not visualizable, falsifiable, and grounded.
Production-shaped static brief + typed hook set for the strongest angle, with a what-to-test-first call.
2 personas + 3 angles; 'cost confession' hook family became the winning test cell
Review mining surfaced a verbatim ('finally stopped hiding my arms') that became the control ad's hook
Copy audit: 6 of 9 elements failed grounding - page proof gap identified before more spend
The agent now replicates the entire Meta Ads AI Agentic System surface (minus rendering and account access): nine modes covering research, angles, the Big-Idea UMP/UMS pipeline, static brief batches, video scenarios with an adversarial roast, live Meta Ad Library competitor spy, the 7-step funnel diagnosis, launch planning, and copy audits.
New public agent: the Meta Ads AI Agentic System's creative doctrine in one hosted run - psychological personas, angles through the 3-Rule Filter (kill list included), a production-shaped static brief, and Found-Not-Written video hooks from your product URL and reviews.
Each system runs on Claude Code or Codex · operator approval on every action · your data stays yours
4 straight answers about what this agent needs from you, how it does the work, and what you get back when a run finishes.