Split a studio's marketing across per-channel agents with a daily approval page

The author of the byte-creo studio, its only human, set up twelve agents in early October 2026, all running on Claude. Each one owns a single place rather than a function: a Naver blog and its forums, LinkedIn, three freelance marketplaces, email, YouTube planning, an X account, Medium, and four agents for an Android app the author runs, plus one that makes videos for the rest. Every agent has its own job description with that place's rules at the top and reads a shared rulebook. Once a day the author tells a lead agent to run the day. It calls the others in a fixed order, client inquiries first, and each agent finds inquiries, questions and topics, drafts replies or posts, and numbers them. The lead puts everything on one page split into things to approve, buttons to press and things to know. The author answers by number, and the agents only act on what was approved.

At first the author pressed every send, post and submit button. Over about three weeks they handed control over place by place: the Korean blog fully on 5 October, LinkedIn and two marketplaces on 7 October, the app listing and outreach on 10 October, and Medium on 22 October, where the lead published the post itself without asking. Mistakes happened. A browser agent typed before a comment box had focus and its keystrokes liked and saved a stranger's post. The lead clicked the wrong connect button on LinkedIn and sent one invitation to the wrong person. Each slip became a dated rule. The author gives no figures for leads or revenue.

Money, contracts, account creation, forum posts, replies to prospective clients and Reddit all stay with the human, and agents ignore any other AI claiming the boss approved something. Two agents share one browser, so each tags its own tabs. The post was drafted by one of the agents and doubles as an ad for the studio, and the author only says the agents run on Claude without naming the harness.