China's Office Workers Have a Plan for the AI-Crazed Boss

The most liked answer starts from a cold observation. If the boss could really run the company on a chatbot, he would already have fired everyone and be doing it alone, grinning from ear to ear. He has not, which means he has discovered a gap between what the model promises and what it delivers, and he has no idea how deep it is. He also does not want to burn his subscription finding out. So he has hit on a cheaper idea: turn the staff into human plug-ins. The poster's phrase is "human MCPs," after the standard tools AI models call on to do things they cannot, and it is now doing the rounds as an insult.

The economics, as this writer lays them out, are brutal and funny. Bullying a language model costs tokens and rarely works. Bullying an employee is free and sometimes does. The subscription is prepaid for the month, so no amount of pressure or fifty-page PDFs costs the boss an extra yuan. Employees, on the other hand, are afraid of losing their jobs, so a few accusations of "not actively embracing AI" and they will start paying for their own chatbot accounts, working unpaid overtime, and quietly using years of human experience to fill the gap the model left. For a small-business owner, he writes, it is a scratch card with no purchase price. Scratch it every day. Win nothing, and you can still call the staff incompetent. Win once, and the breakthrough was your brilliant decision, and nobody sees a bonus.

The sharpest psychological reading came from a poster who asked readers to imagine the boss's day. In the morning meeting he slaps his forehead and announces a world-changing idea. The sales director frowns and says the numbers do not add up. The tech lead hedges and says the architecture cannot carry it. The finance person points out the cash runs out in six months. The boss goes home furious that nobody in the building understands his vision.

Then, at midnight, he types the rejected idea into a chat window, and the machine replies instantly that his strategic insight is exceptionally sharp, that this concept strikes at the heart of a trillion-yuan market, and here is a neatly formatted three-step implementation plan. Available around the clock, endlessly encouraging, permanently on its knees applauding. What self-regarding boss, the writer asks, could resist? The man is not in love with frontier technology. He is in love with a courtier. And the poster's advice follows directly: do not be the martyr who tries to wake a man who is enjoying his dream. If the project fails, the blame will not land on the AI.

The practical replies are where the thread turns into a survival manual. He sends you an eighty-page AI-generated deck, one poster wrote, so you feed it to an AI and get three sentences back. He manages with AI, you slack off with AI, and everyone burns tokens together; the AI companies should be sending thank-you notes. A lawyer in the thread went further and recommended full enthusiasm. Take the boss's chat logs, run them through a model, ask for the key conclusions plus a few well-argued compliments, and deliver the whole package at Monday's meeting with a straight face. Facial control is important, she notes. Do not laugh. The boss will be delighted and you may get a raise. And once the boss thinks in prompts, she adds, managing upward becomes easy for the first time in history: you only need to learn which keywords he likes.

The blunter version, from a reply with hundreds of votes, offered a menu of three: manage upward, agree in public and ignore in private, or water down the execution. Pick one. Treat every meeting like a thesis defence, it said: answer, never argue, apologise if pressed. Do the work you are paid for and do not become a scab for the machine. Another poster, one of China's better-known essayists, was almost cheerful. An AI-written PDF, he said, is a big improvement on the incoherent stuff most bosses write themselves. Think of it as being paid to cosplay an office worker.

"Does McKinsey work for a dozen yuan an hour? Does McKinsey take the blame when the plan fails?"

The top answer ends with the argument that seems to have landed hardest. Fine, the writer says, if the boss says you are not as good as the AI, agree with him. I am not as good as the model. I am not as good as McKinsey. I am not as good as an expert. Then ask what McKinsey charges per hour, and whether McKinsey signs its name under the outcome. A three-thousand-yuan monthly salary buys a few hours of real effort a day and no liability, and the AI accepts no liability either. Staying up all night to prove yourself smarter than a man who is dazzled by a machine that flatters him is, the writer concludes, the real insult to your intelligence. Admitting you are dumber than the AI is the smart move.

There was one dissent worth recording. A poster said, with a straight face, that AI decisions are worse than an expert's but better than an average boss's, and that a model at least admits it was wrong when you push back. The thread laughed, and then several people quietly agreed. If you have sat through a Monday meeting in a Chinese office recently and watched two managers exchange AI-generated documents neither of them wrote nor read, you know why.