We build content with pipelines and then deliver it by hand. This brief states the problem, shows the data, and hands you the research. How you solve it is up to you.
Our fulcrum pipelines now do real work: research floods the board with ideas, the production system writes slides, captions, and SEO strings and designs the visuals, the Dream 100 system runs whole engagement loops. But every single one of those pipelines ends the same way: a finished asset sitting in Canva or Notion, waiting for a person to hand-carry it to Instagram.
Nothing we have schedules or publishes. The last mile of every content workflow is a human pasting, uploading, and clicking — every workday.
From our own EOD logs, May 29 – July 17 (7 weeks) — tasks we ourselves labeled "No AI on this":
"Scheduled and manually posted content" appears in our EODs nearly every workday, from all three of you. It is the single largest block of automatable manual work in the company — bigger than any other category we measured.
Metricool — the scheduler we already pay for — ships an official hosted MCP connector (ai.metricool.com/mcp) that works with Claude Code, Claude Desktop, and Cowork-class clients. Confirmed from Metricool's own docs:
Which means the bridge is probably not an infrastructure project. It's a skill — the shape you already know how to build: something that reads our Notion content board (where posting dates and approval statuses already live) and turns "Approved" into "Scheduled."
| Option | What it is | Trade-off |
|---|---|---|
| Drive Metricool's planner in the browser | A skill that operates the Metricool web planner directly — the same pattern the production orchestrator already uses to drive Claude Design. | Plan-proof and familiar, but more brittle than a connector. |
| Meta's official Graph API | Free, direct publishing of posts, carousels, and reels (up to 25 API posts per account per day). | Real setup cost: a Meta app, tokens per client account, and stories go through a separate, limited API. |
| Postiz (open source) | A 29k-star self-hosted scheduler supporting IG and 30+ platforms. | New infrastructure to run, and it duplicates a tool we already pay for. |
Automating the posting makes the final human quality check more important, not less — once this works, a mistake ships to a client's audience at machine speed. Whatever you build must make the QA pass structural, not optional. Ideas to work with (yours to improve on):
This is a high-agency build. This brief deliberately contains no step-by-step. You have the problem, the data, the verified front door, the fallbacks, and the guardrails. Design the solution: what the skill (or chain of skills) looks like, how it plugs into the pipelines we already run, where the approve gate sits, and how we'll know it worked.
Done looks like: a week's QA'd content goes from our Notion board into Metricool's queue without anyone hand-carrying a single post — demoed live on the Illume account first, with the receipt-vs-reality check passing for a full week after.
Whoever takes this closes the biggest gap in the whole system, for every client at once. Bring it to the weekly session when you have something working — or questions worth arguing about.