I post to social media every day, and most days I write nothing from scratch. AI drafts, adapts, and schedules. But nothing — genuinely nothing — goes public without a human tick. That one design decision is why the system still exists. Here’s the shape of it.
The pipeline: draft → review → publish
Three stages, deliberately separated.
Draft: every morning, AI turns raw material (notes, things I learned, ideas captured during the week) into post drafts, adapted per platform — different length and tone for each.
Review: drafts land in a queue I can read in two minutes. I tick the ones that ship, edit or delete the rest. This is the human gate.
Publish: at posting time, a script publishes only the ticked items through each platform’s official API. Not ticked, not posted. No exceptions in code, so no exceptions in practice.
Why the gate is structural, not a promise
The tempting version is “AI, please only post good things.” It doesn’t work. AI drafts confidently wrong claims, weird tone, and the occasional sentence you’d never say. If the gate is an instruction, it eventually gets skipped. If the gate is structural — the publisher literally cannot see unticked drafts — it never gets skipped. Build the safety into the plumbing, not the prompt.
Use official APIs, not browser bots
Tools that fake being you in a browser break weekly and can get accounts flagged. Every platform worth posting on has an official API. It’s more setup once, then it’s boring — and boring is the goal.
The unglamorous realities
Access tokens expire — usually every ~60 days on Meta platforms — so put the renewal date in your calendar or your pipeline dies silently on a Tuesday. Log everything the system posts, because “did that go out?” is a question you will ask. And prepare content in batches, since AI is much better at making post #7 when it can see posts #1–6.
What this buys you
Consistency without becoming a content zombie. The machine handles “post every day.” You handle “is this true, and does it sound like me?” — which was always the only part worth a human’s time.