A demo proves the model can do it. A workflow proves the team will.

Muse is an agent-based content workflow. I picked it up and loaded it with the things that actually decide whether the output is usable: brand voice, audience, ICP, banned words and the competitor set. Then I built two new agents and four skills on top of it, and worked with engineering to hook it into APIs like Google Analytics and HubSpot.

The part that makes it a system rather than a tool is the instrumentation. Every piece of content routes to a hero page that captures emails.

Data input also means that content stops being shipped blind.

Two marketing people run it, on different companies. A month of campaign content takes 20 minutes. It has been running without me since I handed it over, which is the only test of an AI workflow that counts.