Seven states. One loop. It feeds itself.
Every post a brand publishes moves through the same pipeline — and the engagement it earns becomes the raw material for the next idea. Here's each turn of the loop.
- 01
Capture
Ideas arrive from anywhere — a Telegram drop, the web form, or the engine itself proposing angles off your competitors' best-performing posts and live opportunity signal: rising keywords, trends, and breaking catalysts. Each lands in the inbox, scoped to one client brand.
- 02
Iterate
Generate 3–5 alternative angles in the brand's voice (pinned to Claude Sonnet 4.5). The voice profile, ground-truth facts, and available assets are injected — and the exact prompt is recorded with the output.
- 03
Score
Rate iterations 1–5. A percentile-gated ranker learns which hooks, formats, and angles actually earn engagement for this brand, and biases future generation toward them.
- 04
Promote
Pick the winner. Provenance — idea, iteration, model, and prompt — is copied onto the post, so six months later you can always answer “why did we say this?” without a fragile join.
- 05
Review
The default gate. Out of the box a human — or the client, via batch approval — signs off before anything publishes: the safeguard against shipping a typo across four brands at 7am. Once a brand has earned your trust, autonomous posting can be turned on per brand and per channel; the gate is a policy you control, not a wall.
- 06
Publish
Ship across the v1 channels — LinkedIn, Meta, and YouTube. Each platform is its own render with its own metrics keyed to it, so performance never gets blurred into one fuzzy average. TikTok and more channels arrive in v2.
- 07
Learn
Engagement and cross-channel metrics flow back as signals — re-weighting the ranker and seeding the next capture. The last state feeds the first. The loop closes, and gets sharper every turn.
See it on one of your clients.
A 20-minute walkthrough, capture to published, on a brand you actually run.