Monday night, staring at a dashboard that felt stranger by the hour
Last Monday night at 10 p.m., I was in Hangzhou, updating the pricing page for a freelance project. The client had sent six requests over the previous week, and I had let AI make each change quickly. Then I reopened the dashboard and suddenly couldn’t tell which part of the code handled which feature. I’ve been there too: the more comfortable I became with the tools, the less willing I was to shut down my laptop. The pages appeared to work, but I couldn’t explain why they worked that way.
This isn’t a problem limited to a few people. The more a team relies on AI, the more frequent the code changes become, while the time people spend reading and verifying those changes doesn’t rise at the same pace. It’s like footage that changes five times per second: if I only sample five frames per second, I’ll miss details when I play it back. The same applies beyond programmers. When AI edits my website, drafts copy, or adjusts a workflow, I still need to make room to inspect the work myself. Otherwise, what I deliver gradually slips out of my control.
The rule, and people already using it
My friend Chen Qi runs a one-person company in Chengdu, building reservation pages for restaurants. She used to accept eight revision requests a day and only check the result before delivery. Then she started opening every revised page herself, checking the mobile display and payment flow, and asking the client to test it before sign-off. She also kept a plain document titled “What AI Changed” and “Why I Approved It.” It took a little longer, but she could explain every revision and spent less time fixing the same problem again.
The real reminder isn’t to use AI less; it’s not to hand over both understanding and verification. I keep it simple: multiply the number of AI changes by 2, and that’s the minimum number of times I should check. It’s a prompt, not an exact formula. If a change touches customer records, pricing, or payments, I’d check it once more.
Try it today: $0, 30 minutes, no technical barrier
I can replicate this for $0 and 30 minutes. The only technical requirement is knowing how to open a web page or document. First step: create a plain document called “AI Change Log.” If AI is editing a website, quote sheet, or client workflow today, open the file or page it just changed and write down what changed. There’s no need to build a complicated system yet; completing one check is enough to begin.
What I’d do at each stage
Just starting: I’d include review time in my quote instead of hiding the checking time behind “AI did it in three minutes.” With 1–2 clients: I’d leave a short note for every revision and ask the client to walk through the key steps before delivery. Scaling up: I’d turn recurring changes into a checklist that design and operations staff can also understand, then consider a more formal management tool once the team has stabilized. Not everyone needs this process, and I wouldn’t overhaul everything tonight. It’s fine not to try it yet; simply reviewing the next AI change is already worth something.