Build your first AI content pipeline in one afternoon

A pipeline is not a folder full of prompts. It's an order of operations that produces the same quality whether or not you feel inspired today.

Most people build one backwards. They start at generation — open the tool, type something, see what happens — and then try to bolt a process around the accident. What they end up with is a slot machine with a schedule attached.

The order that works starts further upstream.

Stage 1 — Capture the idea as a tension, not a topic

"AI for creators" is a topic. It cannot produce a good asset, because there is nothing in it to disagree with.

A tension looks like this: creators post three times a week and still feel invisible, because nothing they make looks like them. Now there's a person, a cost, and a cause.

Store three things for every idea: the tension, who feels it, and the evidence you have. If you have no evidence, the idea isn't ready — it's a hunch wearing a headline.

Stage 2 — Write the brief

Six fields, one file, four minutes:

WHO IT'S FOR   one specific person
THE ONE IDEA   a sentence someone could disagree with
THE LOOK       five adjectives + two reference files
THE FORMAT     the container (30s video / 7-slide carousel / 150-word post)
THE NO-LIST    three things this never does
SUCCESS TEST   how you'll know it worked, before you make it

This is where quality is actually decided. Everything downstream is execution, and execution is cheap. Skip this and you'll spend the afternoon generating variations of nothing.

Stage 3 — Generate the raw material, on a timer

Six to eight images, or one script, or both. Change exactly one variable per round so you learn which one is doing the work.

Set a thirty-minute timer. This is the single most useful sentence in this note. Generation will happily expand to fill your entire afternoon, and the marginal image is worth almost nothing. The twentieth variant does not improve on the eighth.

Stage 4 — Run the kill round

Score every candidate 1–5 on three things: specificity (could this be about any topic?), tension (is something at stake?), and credibility (would your audience believe it from you?).

Delete everything under twelve out of fifteen. Not "keep for later" — delete. The folder of maybe-someday is where momentum goes to die, and you will never open it again.

Stage 5 — Assemble one master asset

One master per idea. The master is the thing that carries the idea — usually the video or the carousel. Everything else is a derivative of it.

Have a template. Same margins, same type sizes, same safe zones, every time. This is the boring part, and boring is the point: you're buying speed on the days you don't have energy.

Stage 6 — Distribute natively

Re-argue the idea for each platform. No sentence may appear in two versions — if you can't write four different sentences, you don't have four posts, you have one post and three crops.

What to automate, and what to keep

Automate retrieval, resizing, transcription, caption variants and scheduling. These are repetitions of a decision you already made.

Keep manual: the brief, the kill round, and the final edit. These are the decisions. Handing them to a model is how you end up with a feed that is technically consistent and completely anonymous.

The afternoon, scheduled

If you want the prompts used at each stage, they're in the prompt index — start with "One idea, seven platforms" and "Carousel architect".