Last Wednesday around midnight, I was revising a draft on an AI writing assistant — 3 versions in, I wanted to throw my keyboard.

How the "super smart" AI became my daily frustration

I used to think I was the problem — I just didn't know how to use it properly. Then last month I read an essay called "When Genius Fails." It nailed a truth I couldn't see: most engineers at AI labs come from elite schools and mostly talk to each other. So what they build increasingly shows off how smart they are — not how well it finishes your work.

My partner Xiaolin (runs an independent consulting practice) hit the exact same wall last month. He tried using AI to clean up client meeting notes. The tool couldn't hold context, and the export format clients couldn't even open. He ended up manually copy-pasting and reformatting — didn't finish until 2 AM.

What the article says + who's already using this thinking

The author is Tyler Alterman (I checked — 20-year tech writing veteran). Three core points:

  • AI labs have "intellectual arrogance" — they assume users can keep up with their logic.
  • Models keep getting more complex, but the entry points you can actually use keep getting narrower.
  • The product that actually helps you isn't the "smartest" one — it's the one that "knows your workflow."

Three friends of mine already work this way: Ajie creates AI images for Xiaohongshu (Little Red Book) and switched back to the simplest version. A-Ming, a lawyer, drafts contracts with AI — he deliberately picks the "dumb" template tools. Susu, who builds courses, has her assistant outline content by hand; AI only does the final polish. Common thread: don't chase "most powerful" — chase "best fit."

3 things you can try in 30 minutes tonight

Money: $0. Time: 30 minutes. Skill level: if you can use a phone app, you're set.

Step 1: Open the AI tool you've been using lately, find "Settings" or "Preferences," and turn OFF everything labeled "Advanced features" or "Auto-optimize." Leave only the basic input box.

Step 2: Pick 3 real tasks from last week (like "write a WeChat Moments post for a client") and redo them in basic mode. Compare — is it actually faster?

Step 3: Write down the parts AI can't do (sending invoices, replying to DMs) into your own "must-be-human" checklist.

Note: this isn't about going back to the Stone Age, and not every AI tool is arrogant. It's about taking back control — you use AI, AI doesn't use you.

Advice for different stages

If you're just starting / side hustle still in trial mode: don't chase "strongest AI." Pick the cheapest one that doesn't lag on you. I use the free version myself. Wait until monthly income hits $700+ before considering an upgrade.

If you have 1–2 stable clients: the key isn't switching tools — it's writing down which parts AI handles and which parts humans do. My partner Xiaolin spends 15 minutes every Monday on this list. His efficiency literally doubled.

If you're scaling up / thinking of hiring: don't buy the "family bucket" bundle. Let your assistant use it for a week first, log where they get stuck, then buy small tools that solve those 3 specific pain points. Save around 60% of budget monthly.

One last thing: not everyone needs this tool right now. Skipping for now is fine — finishing the work in front of you matters most.