Scene Hook

Last Tuesday at 11 p.m., I was still sitting in the living room staring at AI, adding one more line after another: “Make it shorter,” “You missed the quote.” I’ve been stuck at this step too. Even with the tool, I was still chained to the chat box.

What This Method Is + Who’s Already Using It

What people have been talking about lately isn’t “how to write one magical prompt.” It’s “how to let AI run through a small process on its own.” I think of it like this: set the goal first, then let AI check itself, ask follow-up questions, and continue to the next step. For example, my friend Li Xuan was in a co-working office in Nanshan, Shenzhen, at 3 p.m. on Thursday, using this approach to organize client interviews: first AI pulled out the key points, then if anything was incomplete it automatically asked follow-up questions, and finally it generated a client-ready summary. She didn’t have to keep watching the whole time.

What It Costs to Replicate Today

Money: RMB 0-99/month; an AI tool you already use is enough to start. Time: 30-60 minutes to put together the first version. Technical barrier: no coding needed; it’s really just setting up “if this isn’t finished, check again and continue to the next step.” First step: open “New Project” or “New Automation” in the AI tool you already use, then write one fixed goal, like “Turn a client voice note into a 3-part summary; if information is missing, list the questions first.” This isn’t something everyone needs. If I were only asking AI two or three times a day right now, I’d probably skip it for the moment too.

Advice by Stage

If I were just starting out, I’d test it on one repeated task first: weekly updates, quote explanations, interview整理. I wouldn’t jump straight into full automation. I messed that up before too—I wanted everything in one shot, and the more I configured, the messier it got.
If I already had 1-2 clients, I’d hand the “check for omissions” step to AI—for example, whether delivery timing, pricing, or action items were missing.
If I were scaling up, I’d start building trigger-based mini workflows: when there’s a new form, a new recording, or a new client message, AI creates the first draft automatically, and a person gives the final check. That way a human is still in the loop, but we’re not starting from zero every single time.