Blog Autopilot

Blog Autopilot is an optional add-on parented on Blog. It plans and drafts articles on a schedule; a human still approves every one before it goes live — the same Approve for publish gate that applies to any AI-authored draft in Blog.

How it works

A content plan sets a cadence (a recurrence rule, in a timezone) and a topic configuration. The scheduler checks for due plans and generates briefs — the next batch of topics to write.
Generation claims briefs and drafts full articles from them, landing as draft with AI generated set — exactly like MCP-authored drafts.
A human edits and approves. Whatever changes between the AI draft and the published (or reviewed) version is captured as edit feedback (before/after text, tied to the event that produced it).
Distillation turns that accumulated edit feedback into learning signals — the patterns your editors actually apply — so future drafts trend toward your house style instead of repeating the same corrections.
Retention keeps this loop from growing unbounded: raw edit text is kept 90 days, feedback that hasn’t been distilled yet is capped at 30 days, and superseded signals are purged after 180.

Guardrails

Monthly article cap and monthly cost cap — Autopilot stops generating once either limit is hit for the month.
Require-review override — a store can force human review even where the base Blog config might otherwise allow more autonomy.
Empty-queue notification — if a plan has gone quiet (no briefs to work) for about a week, the users on that plan’s notify list get told, instead of the pipeline silently doing nothing.

Admin

Content → Blog → Autopilot: content plans, a plan editor (cadence, topic config), run history, and the accumulated learning signals.

Configuration

Autopilot lives entirely under Blog’s own permissions and settings — there’s no separate credential to wire up beyond the LLM routing your deployment already has configured for AI features.