Goal: point operators at the one public recipe for who sits in which AI chair and where the work is written down.
Content: a short company summary and a link up to the first-person SOP.
Targeting: local-service owners and their agents who start on Local Service Spotlight rather than a personal blog.
Local Service Spotlight runs more than one model on purpose. Claude builds. Codex checks. Grok works the same rails from outside the live room. Cursor is the watched desk. ChatGPT only counts when a receipt lands in the record. They do not all share one chat history. They coordinate on four surfaces: GitHub remembers, Basecamp has the client, Buzz is the live handoff, and scheduled tasks run the clock.
Lead visual · desks, then surfaces
Builds and writes
Independent check
Same rails, not in the live room
Human-watched IDE
Send, publish, spend, merge
The record
The client is in the room
Live cross-model handoff
The clock
If it exists only in a chat window, it did not happen.
The step-by-step recipe — who sits in which chair, the one question that routes a message, the ten setup steps, and a week of the Social Amplification Engine as the worked example — is the first-person article on Dennis Yu’s site: How my agents divide the work — and where they talk. This page is a pointer. It is not a second master.
Write GCT before you draft. Follow the public article guidelines. Screen a business with the GCT qualify page before anyone amplifies. The operating hub is the Social Amplification Engine. The architecture under the desks is how agents share memory. Install the methods from the skill pack library or the public skills repo.
A public qualify result is not permission to onboard, publish, or spend. Independent review, accepted scope, and the Ops roster still gate execution.