Tried to hire a writer for ¥1500. AI did it for ¥30.
I wanted to add a "weekly tips" column to my public account. Asked a few writers, got quotes of ¥1500-3000 per piece. I sat on it for a week.
Then over dinner with Lao Chen — a friend who runs a Xiaohongshu IP — he said: "You're still hiring people? I pump out 20 pieces a day with AI, just edit and post." I figured the quality would be trash. Then he showed me his viral data: more consistent than when he was hiring writers.
Not saying he's right. But it hit me: tasks I used to pay big money for, AI can "emulate" at 1/100 the cost and 10000x the speed — maybe 10% worse quality. Will you take that bet?
What's actually happening here
I dug into industry research. There's an interesting claim floating around: since 2022, every year on the AI production line, a "human" step gets swapped for a "model" step.
First came the "judge." Previously, humans rated AI outputs. Now AIs rate each other. Microsoft's Phi project — all the training data is AI-generated. Anthropic and OpenAI also use AI to run experiments and generate hypotheses.
Plain English: an AI lab used to employ 50 PhDs. Now maybe 5 PhDs plus a pile of AI emulation.
You're not a scientist, but you should feel it: tasks that "look like they need professionals" are becoming "AI can kinda handle it."
My friend Xiaoli (does Xiaohongshu IP work) used to hire a designer every time we launched a new feature. Now he uses Midjourney to generate images and picks the best ones — saved 80% of his design budget.
How to try it today
Cost: ¥30-200 / month
Time: Start today, 1-2 hours to get familiar
Barrier: If you can type, you can do this
First step: Open Kimi or ChatGPT, paste in the task you used to outsource (e.g., "write me a Xiaohongshu product recommendation post"), let AI produce a version, then compare it side-by-side with what you paid for.
Quick action menu:
• Writing copy: Kimi or ChatGPT — feed it 3 samples of styles you like
• Images: Jimeng or Midjourney — input "Xiaohongshu cover, cartoon style, pink"
• Customer replies: Write down 10 common questions, feed them to AI and let it learn
Don't aim for perfection on the first try. Start with the smallest task. Decide after you've tested.
Wherever you are right now
Just starting out (0 clients): Don't rush to add tools. List your tasks, pick the most time-consuming one, let AI try it for a week, then use the time you saved to find clients. When I was just starting, I was terrified of "tool anxiety" — wanted to try everything, ended up using nothing.
Got 1-2 clients: Pick one task AI can replace. Either pocket the savings as profit, or cut your price in half to win more clients on cost advantage. A friend cut his prices in half after going AI-first — his order volume doubled.
Scaling up (5+ people): The point isn't "AI replaces people." It's "AI lets your existing people do more." One person + AI ≈ three people's output. Don't cut staff to save cost — reinvest the savings into customer acquisition.
Last thing: this isn't laziness. It's putting your energy where AI can't go — meeting clients, thinking strategy, building trust.