Operations
30 playbooks, run from written steps
Nhan Dinh leads Seller Labs, which makes tools for Amazon sellers. He writes that he runs more than 30 playbooks this way: the weekly revenue sheet, paying recurring vendor invoices, and publishing posts like the one he wrote. Each playbook is a written set of steps, like a standard procedure for a human assistant. On October 5, 2026, he moved his AI work to a newer model by changing one model name in a config file. He rewrote no playbook to do it.
How it wentThe result he reports is the system itself: more than 30 repeated procedures run from written steps, and a model swap that needed no rewrite. He gives no time or money figure. His advice is plain: pick one procedure your team runs every week, not the hardest one, and write its steps down first.
Try it · based on the source's workflow
Turn this weekly procedure into a playbook an AI can run: [procedure name]. Write the steps in plain language, in order, with the exact source for each input: [files, tools, accounts]. Mark every step that sends, pays or deletes as stop-and-ask: show me the finished work first and wait for my approval. Run it once on last week's data and show me the output next to the result I did by hand, so I can check it step by step.
A starting template based on the workflow the source describes. Review it before you use it; you are responsible for what it does.