Lab notes

Every workflow here is built around one rule: AI can draft, score, summarize, and route work, but a human approves anything that publishes, sends, or replies.

That isn't a limitation bolted on at the end — the approval gate is part of the workflow design. The system prepares the work and stops; you give the yes. It's the difference between automation that scales your judgment and automation that replaces it.

For anything touching audience trust — publishing, sponsor outreach, factual claims, voice-sensitive drafts — the default is always: hold for human review.

The Newsletter Research Desk is the proof anchor: the first workflow packaged into something you can actually import, not just read. It's live now as an importable beta (v0.1.0).

The package ships as a portable core — an n8n workflow JSON with placeholder credentials, a Notion/Airtable schema for scored sources, sample input and output, an approval checklist, documented failure modes, and a setup guide with a real time estimate.

The other workflows stay honest in the meantime: they're marked recipe previews until they're packaged the same way.

A creator voice profile captures the things a model can't guess: your point of view, the phrases you use and avoid, your structure preferences, and the lines you won't cross.

Done well, it's reusable — the same profile feeds your newsletter, your repurposed posts, and your outreach, so AI drafts start in your voice instead of a generic one.

It's also a living document. Update it as your positioning sharpens, and keep the judgment calls — what's on-brand, what's too far — with a human.