Nothing should auto-publish: the case for approval-first social

In the automation forums, people who have wired AI content pipelines together are all improvising the same missing piece: an approval step. Webhook emails. Airtable views. A custom dashboard someone's cousin built. The instinct is right — unreviewed AI content shouldn't ship — but the duct tape keeps failing in the same places.

Where DIY approval breaks

Email approvals lose the thread the moment a client replies "can we change the second line?". Spreadsheet gates show the copy but not the post — a caption that reads fine in a cell can die inside a 4:5 crop with a headline overlay. And none of the improvised flows keep a usable record of who approved what, which matters more every month (see our piece on the EU AI Act's Article 50).

Meanwhile the brewery thread shows the inverse failure: owners posting unreviewed AI output directly, engagement sliding, and staff powerless to hold the line. When there is no gate, the feed becomes the gate — and it grades harshly.

What a proper approval flow looks like

• Platform-accurate previews — the client approves the post as it will appear, not a paragraph in an email.
• One place to comment — feedback attached to the asset, not scattered across inboxes.
• Rejection that does work — in Brandaris, a rejection carries its feedback straight back into regeneration, so "not quite" produces a new version in the queue rather than a meeting.
• A log — who approved, when, and what the AI's role was. Quality control and audit trail in one motion.

Approval is the product

The strongest version of this argument is commercial, not defensive. The approval step is where your judgement — the thing clients actually pay an agency for — becomes visible. "Fully autonomous posting" is the wrong goal even where it is technically possible; the right goal is a studio where machine speed and human accountability are the same workflow. That is the line we have drawn in Brandaris, permanently: nothing auto-publishes, ever.