The Operator's AI Brief
What You Still Do Yourself
Two creator videos from October 13 give a small business owner an order of work for AI: list the tasks, test them, and decide what stays yours.
Listen to this brief (2 min)
The audio is AI-produced: an AI clone of Jon's voice and an AI co-host, reading a script built from the sources below.
Two YouTube videos posted on October 13 give a small business owner a usable order of work for AI: list the tasks, test them, and decide which ones stay with you. I read each video’s title, description, and chapter list, not the transcripts.
Nick Saraev’s week-long test
On October 13, Nick Saraev posted a video titled “I Let Claude Code Run My $1M/Year Business for a Whole Week.” The title is his claim, and I have not verified it. The chapter list is the useful part. It runs in order: audit your work, list every task, plan the systems, build evals, turn the winners into skills, schedule what runs without you.
The evals chapter is titled “Aim for 8/10.” A small shop can copy that. Take one repeat task, such as quote follow-ups or appointment reminders. Write the steps down. Run the task on a batch of real examples and count how many come out right. At 8 out of 10, put it on a schedule. Below that, fix the steps first.
The second-to-last chapter is “What I Still Do Myself.” My read: decide which jobs stay with you before you hand anything off.
Jeff Su on personal agents
Also on October 13, Jeff Su posted a video explaining personal agents. His description lays out three setups: a chatbot, a do-it-yourself agent, and a personal agent that offers help before you ask. He says the difference comes down to how much your AI remembers and how much setup work you do.
Personal agents are easier to start. Your own workspace gives you more control, with memory files you can move between tools. A chapter covers privacy, and the description names ecosystem lock-in. Those are the questions to answer before you hand over your customer list or your pricing.
Both creators show a method, not measured results. Treat the sequence as a starting checklist and test it on one task.