The content map from Oct 7 is the input — the queries your ICP actually types, ranked by how close they sit to a purchase. Topics come out of a table, not out of whatever occurred to you in the shower.

The agent drafts against a voice file: your vocabulary, your register, the things you never say. That file is a brand asset — it sits next to the design tokens from Oct 7 and it's written out of the positioning brief from Sep 16.

Decide where a human actually stays in the loop. A checkpoint you always rubber-stamp is theatre — remove it or mean it. The honest answer is usually: read the first ten closely, then spot-check.

One piece of work reshaped per channel automatically — the thread, the post, the newsletter section, the short. Different register each time, same claim underneath, no second writing session.

Scheduled ahead at a rate that survives a bad month. Then it comes back around: last week's numbers choose next week's topics, which is what makes this a loop and not a batch you burned a Sunday on.

One campaign, a couple of ad sets, clear names. Complexity buys nothing at this budget and costs you the ability to say what worked. If you can't read the account in thirty seconds, you can't act on it daily.

Five ways to say one thing is one test. Five different reasons to care is five tests. The angles are your value props from Sep 16, one per ad — if you never wrote them down, this is where that bill arrives.

Start where you're already confident the buyer is. Going broad first spends your budget teaching the platform something you could have told it. Widen only after something converts.

One look a day, three options, judged against a threshold you set before spending. The discipline isn't watching more closely — it's deciding in advance what would make you stop.