The brief was simple: write an AI playbook for the client's twelve-person ops team. The kind of thing they could open on a Monday morning, learn how to use Claude or n8n properly, and stop pinging me on Slack every Tuesday afternoon with the same five questions.
We took the brief seriously. Three founders sat with us for two hours each. We mapped every recurring workflow, drafted prompt templates, color-coded the decision trees, embedded screenshots from their actual tools. By day eleven the Notion playbook was 50 pages, professionally written, table-of-contents-and-everything.
Adoption after week one: zero. Not low. Zero. Nobody had opened it.
The pattern was easy to miss because the playbook itself was good. The problem wasn't quality — it was that we'd shipped a thing for the wrong moment. People learn new tools the way they learn new recipes: by watching someone do it once, not by reading a cookbook. The Notion was a cookbook. Their team was hungry on a Tuesday afternoon, not curious on a Sunday morning.
So on the Tuesday I deleted the entire playbook. I opened Loom. I walked through the same five workflows the playbook had documented — but doing them live, with the real tools, with my real voice and my real mistakes. I cut on the second take if the first one had been too clean. Total runtime: 22 minutes.
I dropped the Loom in their Slack with three lines: "Watch this once. Bookmark it. Use it as your reference when you're about to ask me something on Tuesday afternoon." Within four days, eight of twelve team members had watched it end-to-end. Tuesday Slack pings dropped to under one per week within three weeks.
The playbook wasn't wrong. It was inert. A team SOP for using AI tools should look like a senior teammate doing the work over their shoulder, not a manual sitting on a shelf. Written docs are reference material. Recorded walkthroughs are tutorials. They are not interchangeable, and we keep building the wrong one because writing feels more professional than recording.
