Meet Rae: The AI Strategist Who Rewrites Every Script Before It's Shot

Rae isn’t a hire. She’s a BMad agent: a persona defined in a skill file, loaded into Claude Code, that reviews every script in Marguerite’s pipeline before it goes anywhere near a Higgsfield render. Think of her as the strategist who reads the draft before it becomes a paid decision, because a bad note caught in a script costs nothing and a bad note caught in a finished video costs a render.

Her actual first outing: the mermaid teaser’s script went through eight versions in about forty minutes, then got handed to her. She cut it from forty seconds to thirty, moved the hook earlier, killed a second open thread that was splitting attention, and turned the ending into a loop instead of a hard stop. Good notes, delivered in about the time it takes to read the script twice.

What she actually judges, in order

Rae doesn’t rank a video by whether it “feels engaging.” She has a fixed signal order and won’t skip it: 3-second retention and completion rate first (did the hook work, did the payoff land), shares and sends per 1,000 views second (the strongest real distribution signal on every platform), saves and rewatches third, follows per 1,000 views fourth, and comments and likes last, useful mostly as a question bank for the next video. She won’t crown or kill a format on one video — three is the minimum sample, and she says “untested” out loud rather than guessing.

The structure she builds every script around

Hook, tension, payoff, loop. The hook is three things landing in the same frame: a visual already mid-action, a spoken line with no greeting, and on-screen text that says something different from the spoken line so the video still works on mute. Tension is the pain or the stakes, with every word that’s just context cut out. The payoff is the reveal, delivered in the character’s own voice. And it ends on a loop or an open thread, a frame that reads like the beginning again, or a question the next video answers.

The guardrail that actually matters here

Every character in this pipeline is openly labelled as AI, on every platform, on every post. Rae treats that as a constraint to design around, not a handicap to hide: a hook has to be strong because the character is AI, not dependent on anyone mistaking her for a real person. She won’t write a stunt that only works if the label goes unnoticed.

---

Keep reading

• Why I Love the BMad Method: Multi-Agent AI Development Done Right: the framework Rae runs on
• Five Takes for Eight Seconds: What Directing an AI Actor Actually Costs: the production Rae’s first notes shaped
• Meet Marguerite: We Built an AI Influencer to See If She Can Hit 21,000 Followers by Halloween: the character Rae is strategizing for