Can an agent watch you work and write your content for you?

Hypothesis

The hardest part of content isn't writing — it's noticing. We build things all week and post none of it. An agent studio that scans what we (and the creators we admire) actually publish and ship, drafts in our voice, and queues posts for one-tap review should beat willpower.

What we built

An agentic content studio: a weekly timer scans 30+ sources (our own channels plus an inspiration watchlist), an orchestrator routes each piece of work to one of four specialists — creative director, copywriter, strategist, brand router — any of which can draft a post into a review column. Brand voice is injected from our actual voice documents. Publishing goes through an explicit human-gated dialog to the social platforms; approving a draft is deliberately not the same as publishing it. Kanban board, full lineage of which agent did what, image generation included.

Learnings

• "Approve" and "publish" must be different buttons. The moment an agent can publish outward, the human gate stops being UX and becomes governance.
• Never let a status-polling error mark a post as failed — transient API errors latching into terminal states was the biggest review catch.
• Agents drafting from your own recent work produce grounded content; agents drafting from nothing produce LinkedIn slop. The scanner is the product.

Log

• 2026-07-10 — Brand-aware multi-account publishing, sources manager, and settings shipped; runs on the server with a Monday-morning scan timer.
• 2026-07-09 — Deployed to the server; publishing gate live.
• 2026-07-05 — Rebuild started from a frozen 2025 Airtable automation.